Read Anytime Anywhere Easy Ebook Downloads at ebookmeta.

com

Advanced Computing and Systems for Security Volume

14 Lecture Notes in Networks and Systems 242 1st
Edition Rituparna Chaki (Editor)

https://ebookmeta.com/product/advanced-computing-and-
systems-for-security-volume-14-lecture-notes-in-networks-
and-systems-242-1st-edition-rituparna-chaki-editor/

OR CLICK HERE

DOWLOAD EBOOK

Visit and Get More Ebook Downloads Instantly at https://ebookmeta.com

 Recommended digital products (PDF, EPUB, MOBI) that
you can download immediately if you are interested.

Advanced Computing and Systems for Security Volume 13 1st

Edition Rituparna Chaki

https://ebookmeta.com/product/advanced-computing-and-systems-for-
security-volume-13-1st-edition-rituparna-chaki/

ebookmeta.com

Intelligent Computing Proceedings of the 2021 Computing

Conference Volume 2 Lecture Notes in Networks and Systems
284 Kohei Arai (Editor)
https://ebookmeta.com/product/intelligent-computing-proceedings-of-
the-2021-computing-conference-volume-2-lecture-notes-in-networks-and-
systems-284-kohei-arai-editor/
ebookmeta.com

Intelligent Computing Proceedings of the 2021 Computing

Conference Volume 3 Lecture Notes in Networks and Systems
285 Kohei Arai (Editor)
https://ebookmeta.com/product/intelligent-computing-proceedings-of-
the-2021-computing-conference-volume-3-lecture-notes-in-networks-and-
systems-285-kohei-arai-editor/
ebookmeta.com

Assessment of Advanced Solid State Lighting 1st Edition

National Research Council

https://ebookmeta.com/product/assessment-of-advanced-solid-state-
lighting-1st-edition-national-research-council/

ebookmeta.com
Economics, 14th Global Edition Michael Parkin

https://ebookmeta.com/product/economics-14th-global-edition-michael-
parkin/

ebookmeta.com

ISE Introduction to Managerial Accounting, 9e 9th Edition

Eric W. Noreen

https://ebookmeta.com/product/ise-introduction-to-managerial-
accounting-9e-9th-edition-eric-w-noreen/

ebookmeta.com

Fixer Upper Heart of a Wounded Hero 1st Edition Cassie

Mint

https://ebookmeta.com/product/fixer-upper-heart-of-a-wounded-hero-1st-
edition-cassie-mint-2/

ebookmeta.com

97 Things Every UX Practitioner Should Know: Collective

Wisdom from the Experts 1st Edition Daniel Berlin

https://ebookmeta.com/product/97-things-every-ux-practitioner-should-
know-collective-wisdom-from-the-experts-1st-edition-daniel-berlin/

ebookmeta.com

Empowerment Series: Becoming An Effective Policy Advocate

8th Edition Bruce S. Jansson

https://ebookmeta.com/product/empowerment-series-becoming-an-
effective-policy-advocate-8th-edition-bruce-s-jansson/

ebookmeta.com
African Special Economic Zones: Lessons and Investments
from China 1st Edition Bryan Robinson

https://ebookmeta.com/product/african-special-economic-zones-lessons-
and-investments-from-china-1st-edition-bryan-robinson/

ebookmeta.com
Lecture Notes in Networks and Systems 242

Rituparna Chaki
Nabendu Chaki
Agostino Cortesi
Khalid Saeed Editors

Advanced
Computing
and Systems
for Security:
Volume 14
Lecture Notes in Networks and Systems

Volume 242

Series Editor
Janusz Kacprzyk, Systems Research Institute, Polish Academy of Sciences,
Warsaw, Poland

Advisory Editors
Fernando Gomide, Department of Computer Engineering and Automation—DCA,
School of Electrical and Computer Engineering—FEEC, University of Campinas—
UNICAMP, São Paulo, Brazil
Okyay Kaynak, Department of Electrical and Electronic Engineering,
Bogazici University, Istanbul, Turkey
Derong Liu, Department of Electrical and Computer Engineering, University
of Illinois at Chicago, Chicago, USA; Institute of Automation, Chinese Academy
of Sciences, Beijing, China
Witold Pedrycz, Department of Electrical and Computer Engineering, University of
Alberta, Alberta, Canada; Systems Research Institute, Polish Academy of
Sciences, Warsaw, Poland
Marios M. Polycarpou, Department of Electrical and Computer Engineering,
KIOS Research Center for Intelligent Systems and Networks, University of Cyprus,
Nicosia, Cyprus
Imre J. Rudas, Óbuda University, Budapest, Hungary
Jun Wang, Department of Computer Science, City University of Hong Kong,
Kowloon, Hong Kong
The series “Lecture Notes in Networks and Systems” publishes the latest
developments in Networks and Systems—quickly, informally and with high quality.
Original research reported in proceedings and post-proceedings represents the core
of LNNS.
Volumes published in LNNS embrace all aspects and subfields of, as well as new
challenges in, Networks and Systems.
The series contains proceedings and edited volumes in systems and networks,
spanning the areas of Cyber-Physical Systems, Autonomous Systems, Sensor
Networks, Control Systems, Energy Systems, Automotive Systems, Biological
Systems, Vehicular Networking and Connected Vehicles, Aerospace Systems,
Automation, Manufacturing, Smart Grids, Nonlinear Systems, Power Systems,
Robotics, Social Systems, Economic Systems and other. Of particular value to both
the contributors and the readership are the short publication timeframe and the
world-wide distribution and exposure which enable both a wide and rapid
dissemination of research output.
The series covers the theory, applications, and perspectives on the state of the art
and future developments relevant to systems and networks, decision making, control,
complex processes and related areas, as embedded in the fields of interdisciplinary
and applied sciences, engineering, computer science, physics, economics, social, and
life sciences, as well as the paradigms and methodologies behind them.
Indexed by SCOPUS, INSPEC, WTI Frankfurt eG, zbMATH, SCImago.
All books published in the series are submitted for consideration in Web of Science.

More information about this series at http://www.springer.com/series/15179

Rituparna Chaki · Nabendu Chaki ·
Agostino Cortesi · Khalid Saeed
Editors

Advanced Computing
and Systems for Security:
Volume 14
Editors
Rituparna Chaki Nabendu Chaki
University of Calcutta Department of Computer Science
Kolkata, India and Engineering
University of Calcutta
Agostino Cortesi Kolkata, West Bengal, India
Ca Foscari University
Venice, Italy Khalid Saeed
Bialystok University of Technology
Bialystok, Poland

ISSN 2367-3370 ISSN 2367-3389 (electronic)

Lecture Notes in Networks and Systems
ISBN 978-981-16-4293-7 ISBN 978-981-16-4294-4 (eBook)
https://doi.org/10.1007/978-981-16-4294-4

© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature
Singapore Pte Ltd. 2021
This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether
the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse
of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and
transmission or information storage and retrieval, electronic adaptation, computer software, or by similar
or dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, expressed or implied, with respect to the material contained herein or for any
errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.

This Springer imprint is published by the registered company Springer Nature Singapore Pte Ltd.
The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721,
Singapore
Preface

This book collects the deeply revised version of papers accepted for oral presentation
at the Eighth International Doctoral Symposium on Applied Computation and Secu-
rity Systems (ACSS 2021). ACSS 2021 took place in Kolkata, India, on April 9–
10, 2021. The Doctoral Symposium was organized by the University of Calcutta in
collaboration with Ca Foscari University of Venice, Italy, and Bialystok University
of Technology, Poland.
This unique symposium is aimed specially to facilitate budding researchers in
pursuing their doctoral degree. Each contributed paper was required to have at least
one enrolled Ph.D. student as one of the authors. This has given an opportunity
to each Ph.D. student to express their innovative ideas and to discuss them with a
qualified scientific community of peers.
Over the years, the overall quality of the papers submitted to ACSS has been
improving dramatically, and their subjects reflect and somehow anticipate the
emerging research trends in the area of applied computation and security. In the
call for papers, the following topics of interest related to Applied Computation have
been listed: Security Systems, Software Engineering, Internet of Things, Artificial
Intelligence, Data Science, Computer Vision, and Algorithms.
The editors are greatly indebted to the members of the international program
committee for sharing their expertise and completing their careful review of the
papers in due time. Their reviews have allowed the authors not only to improve their
articles but also to get new hints toward the completion of their Ph.D. thesis.
The dissemination initiatives from Springer have drawn a large number of high-
quality submissions from scholars primarily but not exclusively from India. ACSS
used a double-blind review process and each paper received at least three reviews
either from the PC members or by external reviewers. The reviewers mainly consid-
ered the technical quality and the originality of each paper. As ACSS is a doctoral
symposium, special emphasis was given to assess the clarity of presentation. The
entire process of paper submission, review, and acceptance process was done online.
After carefully considering the reviews, the Program Committee selected only 27
papers for publication out of 45 submissions.

v
vi Preface

We thank the members of Program Committee and Organizing Committee, whose

sincere efforts before and during the symposium have resulted in strong technical
program and in effective discussions. We thank Springer Nature for sponsoring the
best paper award. In particular, we appreciate the initiative from Mr. Aninda Bose and
his colleagues in Springer Nature for their strong support toward publishing this post-
symposium book in the series “Advances in Intelligent Systems and Computing”.
We would also like to thank ACM for the continuous support toward the success
of the symposium. Last but not least, we thank all the authors without whom the
symposium would not have reached up to this standard.
On behalf of the editorial team of ACSS 2021, we sincerely hope this volume will
be beneficial to all its readers and motivate them toward better research works.

Kolkata, India Rituparna Chaki

Kolkata, India Nabendu Chaki
Venezia, Italy Agostino Cortesi
Bialystok, Poland Khalid Saeed
Contents

Security
Parallel Simulation of Cyber-Physical-Systems . . . . . . . . . . . . . . . . . . . . . . . 3
Kamal Das, Amit Gurung, and Rajarshi Ray
Attack Detection Scheme Using Deep Learning Approach for IoT . . . . . . 17
Vikash Kumar, Sidra Kalam, Ayan Kumar Das, and Ditipriya Sinha
An Efficient Authentication Scheme for Mobile Online Social
Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
Munmun Bhattacharya, Sandip Roy, and Samiran Chattopadhyay
GAN-Based Data Generation Approach for IDS: Evaluation
on Decision Tree . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
Sudhir Kumar Pandey, Vikash Kumar, Ditipriya Sinha, and Ayan Kumar Das

Software Engineering
Conceptualizing Re-configurable Business Process:
A Context-Driven Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
Priyanka Chakraborty and Anirban Sarkar
Dcube N N : Tool for Dynamic Design Discovery from Multi-threaded
Applications Using Neural Sequence Models . . . . . . . . . . . . . . . . . . . . . . . . . 75
Srijoni Majumdar, Nachiketa Chatterjee, Partha Pratim Das,
and Amlan Chakrabarti
Construction of Materialized Views in Non-Binary Data Space . . . . . . . . 93
Santanu Roy, Bibekananda Shit, Soumya Sen, and Agostino Cortesi
Dynamic Prioritization of Software Requirements for Incremental
Software Development . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
Mandira Roy, Novarun Deb, Agostino Cortesi, Rituparna Chaki,
and Nabendu Chaki

vii
viii Contents

Systems Biology
A Framework for Translation and Validation of Digital
Microfluidic Protocols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
Pushpita Roy, Ansuman Banerjee, and Bhargab B. Bhattacharya
Disease-Relevant Gene Selection Using Mean Shift Clustering . . . . . . . . . 151
Srirupa Dasgupta, Sharmistha Bhattacharya, Abhinandan Khan,
Anindya Halder, Goutam Saha, and Rajat Kumar Pal
Multiple Fault Identification and Diagnosis in Cross-Referencing
Digital Microfluidic Biochips . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
Sagarika Chowdhury, Kazi Amrin Kabir, Debasis Dhal,
Rajat Kumar Pal, and Goutam Saha
Brain Tumor Detection: A Comparative Study Among Fast Object
Detection Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
Sunita Roy, Sanchari Sen, Ranjan Mehera, Rajat Kumar Pal,
and Samir Kumar Bandyopadhyay
MicroRNA-Based Cancer Classification Using Feature Selection
Wrapper . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
Shib Sankar Bhowmick and Debotosh Bhattacharjee
Editors and Contributors

About the Editors

Rituparna Chaki is a Full Professor in the A K Choudhury School of Information

Technology, University of Calcutta, India since June 2015. She joined academia as
faculty member in the West Bengal University of Technology in 2005. Before that she
has served under Government of India in maintaining industrial production database.
Besides, she has served as a Visiting Professor in the AGH University of Science
& Technology, Cracow, Poland since 2013 for consecutive years. Rituparna did her
Ph.D. from Jadavpur University in 2002. She has been associated in organizing
many conferences in India and abroad as Program Chair, OC Chair or as member
of Technical Program Committee. She has published more than 60 research papers
in reputed journals and peer-reviewed conference proceedings. Her research interest
is primarily in Adhoc networking and its security. She is a professional member of
IEEE and ACM. Currently, Rituparna is the Secretary for ACMW-India.

Nabendu Chaki is a Professor in the Department Computer Science & Engineering,

University of Calcutta, Kolkata, India. He is the Editor in Chief of the Springer Nature
book series on Services and Business Process Reengineering. Besides editing about
40 conference proceedings with Springer, Dr. Chaki has authored 8 text and research
books with CRC Press, Springer Nature, etc. He has published more than 200 Scopus
Indexed research articles in Journals and International conferences. Prof. Chaki has
served as a Visiting Professor in different places including US Naval Postgraduate
School, California, and in different Universities in Italy and Poland. He is the founder
Chair of ACM Professional Chapter in Kolkata and served in that capacity for 3
years since January 2014. He has been active during 2009-2015 towards developing
several international standards in Software Engineering and Service Science as a
Global (GD) member for ISO-IEC.

Agostino Cortesi, Ph.D. is a Full Professor of Computer Science at Ca’ Foscari

University, Venice, Italy, and Dean of the PhD programme in Computer Science.
He has previously served as Department head, and as Vice-Rector for quality

ix
x Editors and Contributors

assessment and institutional affairs. His research interests include programming

languages theory, software engineering, and static analysis techniques, with partic-
ular emphasis on security applications. He serves as coordinator of the H2020 Euro-
pean project “Families_Share” and of the ITALY-INDIA project “Formal Specifica-
tion for Secured Software System”. He published more than 150 papers in interna-
tional journals and proceedings of international conferences. According to Scopus,
his h-index is 19 and his i-10 index is 39. He serves as co-Editor in Chief of the book
series “Services and Business Process Reengineering” edited by Springer-Nature.

Khalid Saeed is a full Professor of Computer Science in the Faculty of Computer

Science at Bialystok University of Technology. He was with Faculty of Mathematics
and Information Sciences at Warsaw University of Technology in 2014-2019. He
was with AGH Krakow in 2008-2014. He received the BSc Degree in Electrical and
Electronics Engineering from Baghdad University in 1976, the MSc and PhD (distin-
guished) Degrees from Wroclaw University of Technology in Poland in 1978 and
1981, respectively. He received his DSc Degree (Habilitation) in Computer Science
from the Polish Academy of Sciences in Warsaw in 2007. He was nominated by the
President of Poland for the title of Professor in 2014. He has published more than 245
publications including 98 journal papers and book chapters, 87 peer reviewed confer-
ence papers, edited 48 books, journals and Conference Proceedings, written 12 text
and reference books (h-index 15 in WoS base and 12 in SCOPUS base). He super-
vised more than 15 PhD and 130 MSc theses from and outside Poland. He gave 50
invited lectures and keynotes in different universities in Canada, China, Colombia,
Czech, Germany, India, Japan, Serbia, Slovakia and South Korea on Biometrics,
Image Processing and Analysis. He received more than 30 academic awards.
Khalid Saeed is a member of more than 15 editorial boards of international journals
and conferences. He is an IEEE Senior Member and has been selected as IEEE Distin-
guished Speaker for 2011-2016. Khalid Saeed is the Editor-in-Chief of International
Journal of Biometrics with Inderscience Publishers.

Contributors

Samir Kumar Bandyopadhyay Department of Computer Science and Engi-

neering, University of Calcutta, Kolkata, West Bengal, India
Ansuman Banerjee Indian Statistical Institute, Kolkata, India
Debotosh Bhattacharjee Department of Computer Science and Engineering,
Jadavpur University, Kolkata, West Bengal, India
Bhargab B. Bhattacharya IIT Kharagpur, Kharagpur, India
Munmun Bhattacharya Department of Information Technology, Jadavpur Univer-
sity, Salt Lake City, Kolkata, India
Editors and Contributors xi

Sharmistha Bhattacharya University of Calcutta, Acharya Prafulla Chandra Roy

Shiksha Prangan, Saltlake, Kolkata, India
Shib Sankar Bhowmick Department of Electronics and Communication Engi-
neering, Heritage Institute of Technology, Kolkata, India
Nabendu Chaki University of Calcutta, Kolkata, India
Rituparna Chaki University of Calcutta, Kolkata, India
Amlan Chakrabarti University of Calcutta, Kharagpur, West Bengal, India
Priyanka Chakraborty Department of Computer Science and Engineering,
National Institute of Technology, Durgapur, West Bengal, India
Nachiketa Chatterjee University of Calcutta, Kharagpur, West Bengal, India
Samiran Chattopadhyay Department of Information Technology, Jadavpur
University, Salt Lake City, Kolkata, India
Sagarika Chowdhury Department of Computer Science and Engineering, Univer-
sity of Calcutta, Kolkata, India
Agostino Cortesi DAIS, Ca’ Fosacari University, Venice, Italy
Ayan Kumar Das Birla Institute of Technology Mesra, Patna, Bihar, India
Kamal Das National Institute of Technology Meghalaya, Shillong, Meghalaya,
India
Srirupa Dasgupta Government College of Engineering and Leather Technology,
Saltlake, Kolkata, India
Novarun Deb Indian Institute of Information Technology, Vadodara (IIIT-V),
Gandhinagar, India
Debasis Dhal Department of Computer Science and Engineering, University of
Calcutta, Kolkata, India
Amit Gurung Martin Luther Christian University, Shillong, Meghalaya, India
Anindya Halder North-Eastern Hill University, Tura Campus, Tura, Meghalaya,
India
Kazi Amrin Kabir Department of Computer Science and Engineering, University
of Calcutta, Kolkata, India
Sidra Kalam Birla Institute of Technology, Mesra, Patna, India
Abhinandan Khan University of Calcutta, Acharya Prafulla Chandra Roy Shiksha
Prangan, Saltlake, Kolkata, India
Vikash Kumar National Institute of Technology Patna, Patna, Bihar, India
xii Editors and Contributors

Rajat Kumar Pal University of Calcutta, Acharya Prafulla Chandra Roy Shiksha
Prangan, Saltlake, Kolkata, India
Srijoni Majumdar Indian Institute of Technology, Kharagpur, West Bengal, India
Ranjan Mehera Business & Solution Consulting, Subex, Inc., Broomfield, CO,
USA
Rajat Kumar Pal Department of Computer Science and Engineering, University
of Calcutta, Kolkata, West Bengal, India
Sudhir Kumar Pandey Lok Nayak Jai Prakash Institute of Technology, Chapra,
Saran, Bihar, India
Partha Pratim Das Indian Institute of Technology, Kharagpur, West Bengal, India
Rajarshi Ray Indian Association for the Cultivation of Science, Kolkata, West
Bengal, India
Mandira Roy University of Calcutta, Kolkata, India
Pushpita Roy Indian Statistical Institute, Kolkata, India;
Calcutta University, Kolkata, India
Sandip Roy Department of Computer Science and Engineering, Asansol Engi-
neering College, Asansol, WB, India
Santanu Roy Future Institute of Engineering and Management, Kolkata, India
Sunita Roy Department of Computer Science and Engineering, University of
Calcutta, Kolkata, West Bengal, India
Goutam Saha Department of Information Technology, North-Eastern Hill Univer-
sity, Umshing Mawkynroh, Shillong, Meghalaya, India
Anirban Sarkar Department of Computer Science and Engineering, National
Institute of Technology, Durgapur, West Bengal, India
Sanchari Sen Department of Computer Science and Engineering, University of
Calcutta, Kolkata, West Bengal, India
Soumya Sen University of Calcutta, Kolkata, India
Bibekananda Shit Future Institute of Engineering and Management, Kolkata, India
Ditipriya Sinha National Institute of Technology Patna, Patna, Bihar, India
Security
Parallel Simulation of
Cyber-Physical-Systems

Kamal Das, Amit Gurung, and Rajarshi Ray

Abstract Model-based design (MBD) in systems engineering is a well-accepted

technique to abstract, analyze, verify, and validate complex systems. In MBD, we
design a mathematical model of the system to virtually execute and test systems via
model simulations to understand the system dynamics better. Computing model sim-
ulations have their challenges: one is to ensure that the simulation trajectory preserves
the model semantics. Besides, computing many simulation trajectories over a long
time-horizon must be time-efficient for rapid response to system engineers. In this
work, we address these challenges in simulating models of Cyber-Physical-Systems
(CPS), particularly systems possessing mixed discrete-continuous dynamics. We
focus on the subclass of CPS’s hybrid-automata models, where jump predicates are
restricted to polygonal constraints and present a numerical simulation engine that can
efficiently compute many random simulations in parallel by exploiting the parallel
computing capability in modern multicore processors. Our simulation engine imple-
ments a lock-free parallel breadth-first-search (BFS) like algorithm and is imple-
mented in the model-checking tool XSpeed. We demonstrate the performance gains
of our simulation engine over SpaceEx and CORA, the modern model checkers and
simulators for affine hybrid systems.

Keywords Hybrid systems · Hybrid automaton · Simulation trajectory · Parallel

simulations

K. Das (B)
National Institute of Technology Meghalaya, Shillong, Meghalaya, India
e-mail: [email protected]
A. Gurung
Martin Luther Christian University, Shillong, Meghalaya, India
R. Ray
Indian Association for the Cultivation of Science, Kolkata, West Bengal, India
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 3
R. Chaki et al. (eds.), Advanced Computing and Systems for Security: Volume 14,
Lecture Notes in Networks and Systems 242,
https://doi.org/10.1007/978-981-16-4294-4_1
4 K. Das et al.

1 Introduction

MBD in systems engineering is a well-accepted technique to abstract, analyze, ver-

ify, and validate complex systems [20, 25, 26]. In MBD, a mathematical model of
the system is designed and used to virtually execute and test systems via model sim-
ulations with the intent of gaining a better insight of the system dynamics. Model
simulation not only serves as a useful tool to assess the dynamics of a complex
system but also enables detection of bugs early in the design and development pro-
cess, without needing a physical prototype of the system in place, therefore sub-
stantially reducing the time, cost, and effort in system-engineering. There are vari-
ous modeling frameworks such as finite-state-machines (FSMs) [8], Petri-Nets [28],
ordinary-differential-equations (ODEs) [9], message-sequence-charts (MSCs) [10],
timed-automata [5], and hybrid-automata [18] to name some and each one is well-
suited to model a particular category of systems. For instance, time-automata are
well-suited to real-time systems whereas Petri-Nets are ideal for concurrent systems
modeling.
CPS is an integration of physical processes with computation [22]. The physical
processes follow continuous dynamics and are monitored and controlled by a discrete
controller implementing a control logic. There is a growing interest in CPS research
since many real-world applications and a large number of Artificial Intelligence
powered systems are cyber-physical in nature. Unmanned aerial vehicles (UAV),
self-driving cars, autonomous robots, planetary landers, and rovers are all examples
of CPS. The mathematical modeling framework that is particularly suited for CPS is
Hybrid Automaton (HA) since it can represent the discrete computation by a state-
transition system and continuous process dynamics by ODEs [4]. Depending on the
type of ODEs, hybrid-automata are broadly classified into linear, affine, or non-linear.
We present the details in the following sections.
Computing simulation trajectories of HA models of CPS poses difficulty due to
(1) the interplay of continuous and discrete dynamics and (2) computations involving
variables taking values in R. Numerical computation of a HA trajectory is compu-
tationally inexpensive in general. However, when precise trajectories at very small
time-steps are necessary over a large time-horizon, its computation may require con-
siderable time. Additionally, a system engineer may wish to simulate a model with
numerous initial conditions over a large time-horizon for the purpose of testing and
robustness evaluation, thus requiring the computation of a large number of long
simulation trajectories. In-efficient sequential computations may therefore have a
very slow response. These observations motivate us to design parallel algorithms for
computing simulations of CPS models. In this work, we propose a multi-threaded
lock-free BFS like algorithm to compute many simulation trajectories of a restrictive
class of hybrid automaton models of CPS, where the transition jump predicates are
polygonal constraints. A multi-threaded implementation exploits the inherent paral-
lel processing power in modern multicore processors. The algorithm is implemented
in the model-checker XSpeed [27], open-source software available in https://gitlab.
com/raj.ray84/XSpeed-plan. We demonstrate our simulation engine’s performance
Parallel Simulation of Cyber-Physical-Systems 5

benefits over SpaceEx [14] and CORA [2], the modern model checkers and simu-
lators for linear and affine hybrid systems.
Related Works:
Tools like HyLAA [7] and Breach [11] implement numerical simulators for linear
hybrid systems whereas C2E2 [12] is a numerical simulator for linear and non-linear
CPS. These tools, however, focus on methods of approximating reachable states of
HA models using finitely many simulations. Breach can additionally monitor the
robustness satisfaction of metric interval temporal logic (MITL) formulas. SpaceEx
[14] and CORA [2] are verification tools that implement HA simulation engines
that can compute random simulations using numerical ODE solvers. None of these
tools exploits the inherent parallelism in the modern multicore processors to accel-
erate computing simulation trajectories in parallel. This is where we contribute by
implementing a parallel simulation engine as part of the model-checker XSpeed.
We organize the paper as follows. The requisite background is discussed in Sect. 2.
The parallel algorithm to compute simulation trajectories is presented in Sect. 3. We
show the algorithm’s performance and validity evaluation in Sect. 4, and we conclude
in Sect. 5.

2 Preliminaries

An HA is a mathematical formalism to model CPS. In an HA, the discrete behavior

is abstracted by a finite state machine, whereas the continuous behavior is abstracted
by ODE assigned to the automaton’s locations. The definition of an HA is

Definition 2.1 [18] A hybrid automaton is a 7-tuple (X , G (V, E), Init, Inv, Flow,
Jump, Assign) where
– X = {x1 , x2 , . . . , xn } is a finite set of continuous variables. The number of
variables in the set is called the dimension of the hybrid automaton. The set
Ẋ = {x˙1 , x˙2 , . . . , x˙n } is the set of variables representing the first derivative of
the respective variables in X . Similarly, the set X  = {x1 , x2 , . . . , xn } is the set
of primed variables representing the reset values of the respective variables in X
after effectuating a discrete transition.
– G (V, E) is a directed multigraph of a finite automaton with the set of locations V
and the set of transition edges E.
– Init and Inv are labeling functions that assign to each location in V, a predicate
with free variables from X . The predicate assigned to a location by Inv and Init
is called the invariant and initial condition of the location respectively. Jump is a
labeling function that assigns to each transition edge e ∈ E, a predicate with free
variables from X .
– Flow is a labeling function that assigns to each location in V, a predicate with free
variables from X ∪ Ẋ .
– Assign is a labeling function that assigns to each transition edge e ∈ E, a predicate
with free variables from the set X ∪ X  .
6 K. Das et al.

Fig. 1 A pictorial representation of a general HA trajectory

The state of an hybrid automaton is a 2-tuple , v, where  ∈ V and v ∈ Rn such that
v satisfies the predicate Inv(), i.e., Inv()[X := v] = tr ue, n being the dimension
of the automaton. The state represents the location  of the HA and an assignment
of values to the variables of the HA, denoted with vector v. The state of an HA may
change either by a timed transition or by a discrete transition. A timed transition
δ
due to δ passage of time can be represented as , v − → , w such that v, v̇ and
w, ẇ satisfies the flow predicate Flow(), i.e., Flow()[X := v, Ẋ := v̇] = tr ue
and Flow()[X := w, Ẋ := ẇ] = tr ue. Note that in a timed transition, the location
of the state  ∈ V remains the same but represents the change in system variables
due to the continuous flow dynamics. A discrete transition can be represented as
e
→ 2 , v  , given that ∃e ∈ E from 1 to 2 for some 1 , 2 ∈ V, such that
1 , v −
Jump(e)[X := v] = tr ue, and Assign(e)[X := v, X  := v  ] = tr ue. We now define
a trajectory of an hybrid automaton.

Definition 2.2 A finite trajectory of an HA is a sequence of states obtained by timed

δ0 e1 δ1
and discrete transitions of a HA, represented as 0 , v0  − → 1 , v1  −
→ 0 , v1  − →
en δn
→ n , vn  −
1 , v2  . . . − → n , vn+1  such that
1. Init(0 )[X := v0 ] = tr ue.
δi ei
2. ∀i ∈ [0, n], i , vi  −
→ i , vi+1  is a timed transition and ∀i ∈ [1, n], i−1 , vi  −
→
i , vi  is a discrete transition of the HA.
δi τ
3. For a timed transition i , vi  −
→ i , vi+1  in the trajectory sequence, i , vi  −
→
τ
i , vi  is also a timed transition in the HA, for any intermediate time τ ∈ [0, δi ].

Definition 2.3 A trajectory state of an HA trajectory beginning from an initial state

0 , v0  is a state , v obtained from a finite sequence of timed and discrete transi-
tions from the initial state.

An exemplary trajectory of a two-dimensional HA with two timed transitions and

one discrete transition is shown pictorially in Fig. 1. The labeled shaded regions depict
the vectors satisfying the corresponding predicate. The curve vi − vi+1 and vi+1 
−
Parallel Simulation of Cyber-Physical-Systems 7

Fig. 2 An HA modeling a Thermostat

vi+2 depicts the trajectory in location 1 and 2 due to timed transition, satisfying
Flow(1 ) and Flow(2 ) respectively. The figure shows a discrete transition due to an
edge e from 1 to 2 . Since the end-point vi+1 of the trajectory in 1 satisfies the
Jump(e) predicate, the transition is enabled. The result of taking the transition is an

update of vi+1 to the new vector vi+1 due to the assignments defined in the Assign(e)
predicate. An important point here is that it is not obligatory to effectuate a discrete
jump when the Jump predicate is satisfied since HA has may transition semantics.
In other words, the may transition semantics in HA says that if a state , v is such
that v satisfies Inv() as well as the Jump(e) predicate for some e ∈ E, then there is
a choice to either take a timed transition or take a discrete transition due to e from
the state , v in the HA.
Definition 2.4 A polygonal constraint over the variables in X = {x1 , x2 , . . . , xn } is
of the form A.x ≤ b, where An×n is a real-valued matrix, xn×1 is a column vector
comprising of the variables x1 , x2 , . . . xn and b ∈ Rn is a real-valued column vec-
tor. The vectors v ∈ Rn satisfying a polygonal constraint defines an n-dimensional
polytope.
An HA, modeling a thermostat is shown in Fig. 2. It has two locations, ON and OFF,
to represent the thermostat’s switched-on and switched-off state, respectively. It is a
two-dimensional HA with variables T and time, representing the temperature and
the elapsed time. The invariant T ≥ 60 in OFF location signifies that during the
switched-off state of the thermostat, the temperature can be 60 or above. Similarly
the invariant T ≤ 70 in the ON location signifies that during the switched-on state
of the thermostat, the temperature can be 70 or below. The Flow in the OFF location
is Ṫ = −k2, time˙ = 1 and signifies that the temperature decreases in a constant rate
k2 while the time progresses during the switched-off state of the thermostat. Simi-
larly, during the switched-on state, the temperature increases following the dynamics
Ṫ = k1(70 − T ) while the time progresses (time˙ = 1) as represented with the Flow
predicate. The Jump predicate on the OFF to ON transition is T ≤ 62 indicating that
the transition may take effect only when the temperature of the thermostat is 62 or
8 K. Das et al.

below. On effectuating the transition, the Assign predicate T  := T , time := time

depicts that the updated value of the temperature and time is the same as its value
before taking the transition. Similarly, the transition from ON to OFF may take effect
only when the temperature is 68 or above and taking the transition does not modify
the value of the variables. Overall, we see that the thermostat’s temperature will vary
in the interval [60, 70] due to the invariant constraints imposed on the two locations.
We now present our contribution in the following section. This HA is an example
where the Jump predicates are polygonal constraints since they form A.x ≤ b.

3 Parallel HA Simulation

This section presents our proposed parallel trajectory simulation algorithm. We first
briefly describe the computation of timed and discrete transitions.
Evaluating Timed Transition: Trajectory-states due to time-transitions are com-
puted numerically using an ODE solver [29]. The present-day ODE solvers can simu-
late first-order linear as well as non-linear ODEs very efficiently. We can improve the
trajectory’s precision by choosing smaller time-steps but at the cost of performance
and memory.
Evaluating Discrete Transition: In the may transition semantics of HA, there
may be infinitely many next trajectory-states possible after taking a discrete transi-
tion. This is due to the non-determinism involved in either choosing or not choosing
to take a transition when an HA state satisfies both the Jump and I nv predicates.
In our algorithm, we follow as soon as possible semantics where a discrete tran-
sition is effectuated as soon as the trajectory satisfies a Jump predicate. There are
two numerical problems involved in the implementation: (1) detecting whether a
trajectory-state , x satisfies the Jump() predicate, and (2) computing the next
trajectory-state  , x   such that Assign(e)[X = x, X  = x  ] is satisfied. If a Jump
predicate is a hyperplane, then successive trajectory points computed at time-step
δ may cross the guard failing to detect an intersection. To deal with this crossover
detection problem, we convert the predicate from a hyperplane to a half-space [17] in
the region opposite to the region containing the trajectory’s initial point. This ensures
crossover detection. In the case of polygonal predicates other than hyperplanes, the
satisfaction is easily checked from the satisfiability of AX − b ≤ 0. As soon as a
trajectory-state is detected to satisfy Jump(e), for some e, we compute the successor
state following Assign(e). The computation of a simulation trajectory due to timed
transition is shown in Algorithm 1. The algorithm terminates as soon as a successor
state due to a discrete transition is found, or all states in the time-horizon have been
computed.
For a chosen point pt and a given time-horizon T , the algorithm computes a simu-
lation trajectory, T race, consisting of a sequence of trajectory-states. The trajectory-
states are computed using an ODE solver according to the location dynamics, dis-
cretized at a fixed time-step, δ. The data structure pt consists of a field τ that holds
the elapsed time in the simulation. For any T race, the successive pt.τ holds the time
Parallel Simulation of Cyber-Physical-Systems 9

at which pt is attained in the model. Therefore, the number of discrete trajectory-

states for a given time-horizon T is computed as N = (T − τ )/δ (see line 2). The
algorithm computes trajectory-states until a point (v) intersects with a transition’s
Jump predicate (Jump(e)) or all the trajectory-states in the given time-horizon have
been computed (see lines 8–17). Line 8 computes the successor trajectory-state using
an ODE solver. This is added to the trace (in line 9) and is tested for intersection
with each transition’s (Loc.trans) jump predicate g (at line 12). If the intersection
is not {∅}, then, a successor state due to discrete transition is evaluated for fur-
ther exploration. The time elapsed, τ is computed as k × δ, k being the number of
trajectory-states computed at a finite time-step δ. This value is appended in the point
p (in line 14) and returned by the routine. When a trajectory does not satisfy any of
the transition’s jump predicate, it returns an empty point p. This indicates that no
new successor states due to discrete transition are present and the trajectory evalua-
tion is complete for the full time-horizon. We now present the algorithm to compute
multiple trajectories in parallel.

Algorithm 1 Simulation of HA from a state pt

1: procedure ODE- sim( pt, ha, T )
2: N = (T − pt.τ )/δ number of trajectory states
3: Loc ← pt.get Location() from ha
4: I nv ← invariant from Loc
5: T race. push( pt); v = pt; v: initial trajectory state
6: k = 1; p = ∅;
7: repeat
8: Compute next trajectory state v use ODE solver
9: T race. push(v)
10: for each t in Loc.trans do
11: g ← J ump(t)
12: if (v ∩ g) != ∅ then
13: p←v
14: p.τ = k × δ time elapsed in this loc
15: done = tr ue;
16: k =k+1
17: until done OR v ∈ / I nv(Loc) OR k ≥ N
18: return p

Parallel Algorithm: The inputs to our algorithm are N —the number of random
simulation trajectories to compute, the HA model to simulate, and the time-horizon T
for simulation. Random vectors N pts ⊂ Rn are obtained such that for any v ∈ N pts,
Init()[X := v] = tr ue, for some  ∈ V. Our algorithm is motivated by the parallel
BFS implemented in the model checkers XSpeed [15] and Spin [19]. The algorithm
maintains a shared data-structure W ait, a list of states of the HA, from which further
trajectory needs to be generated via timed and/or discrete transitions. We use W ait of
size (N × N ) to randomly distribute the trajectory-states among N available threads
for efficient load balancing. However, simultaneous read and write access must be
controlled with semaphores or locks to avoid a race condition with a shared W ait.
10 K. Das et al.

Locking incurs extra overhead in terms of performance. Our algorithm proposes to

have a shared W ait, yet, with no locking overhead by adapting the parallel lock-free
BFS algorithm proposed in [16, 19] for HA simulation. The routine ODE- sim (line
10) computes the numerical simulations of a trajectory-state with a fixed time-step
δ and generates new successor trajectory-state pts, which must satisfy the jump
condition. We read a trajectory-states from the W ait[t] copy at each iteration and
write a successor trajectory-states to the W ait[1 − t] copy. Initially, t is assigned
0, and at each successive iteration, it is reset to 1 − t (see line 17). In this manner,
the write W ait copy transforms to the read W ait copy and vice-versa after each
iteration. The algorithm terminates when no successor trajectory-states remains in
W ait[1 − t] or when the BFS reaches a user-defined depth bound. The algorithm
also terminates when the trajectory-states have elapsed the time-horizon (T).

Algorithm 2 Parallel BFS Algorithm for Parallel Simulations of HA

1: procedure PAR- SIMU(ha, N pts, T )
2: t = level = 0, N = Cor es
3: W ait[2][N ][N ] 2 × N × N vector
4: Distribute N pts equally in W ait[t][i][0], for i = 0 to N − 1
5: repeat
6: Fork threads with id w = 0 to N − 1 and execute
7: for q = 0 to N − 1 do
8: for each s in W ait[t][w][q] do
9: delete s from W ait[t][w][q]
10: pts ← ODE- sim (s, ha, T )
11: w  = select random 0 . . . N − 1
12: add pts to W ait[1 − t][w  ][w]
13: Thread synchronization Threads sync here to ensure BFS
14: if W ait[1 − t] is ∅ then
15: done = true;
16: else
17: t =1−t Read/Write transformation
18: level ← level + 1
19: until done OR level = depth

4 Evaluation

Benchmark Description:
We consider Bouncing-Ball [23], Navigation benchmark [13], Thermostat [3], Heli-
copter Controller [30], Five Dimensional dynamical system [1], Vehicle platoon
[24], Drivetrain [21], and Building [6] benchmarks for evaluation. The Bouncing-
Ball models the motion of a ball under gravity together with bouncing upon hitting
the ground. The navigation-benchmark depicts a moving object in a grid of n × n
partition in a plane. A Thermostat is a model of a temperature controller. The Heli-
Parallel Simulation of Cyber-Physical-Systems 11

Table 1 Performance comparison of computing 1000 random simulations over SpaceEx

Time (in secs) Speedup Vs
Benchmark Dim #Loc SpaceEx XSpeed XSpeed XSpeed SpaceEx
Seq Par Seq
B. Ball 3 1 13.50 1.84 1.29 1.43 10.46
Nav (1) 4 9 11.54 0.95 0.33 2.88 34.97
Nav (2) 4 25 63.64 2.52 0.83 3.04 76.67
Nav (3) 4 81 85.73 3.61 1.15 3.14 74.55
Helicopter 29 1 58.16 18.30 4.13 4.43 14.08
Building 48 1 32.48 73.21 13.39 5.47 2.43

copter controller represents the controller of a Westland Lynx military helicopter.

The Five-dimensional system is a continuous linear dynamical system. The Vehi-
cle platoon models a platoon of three vehicles with a leader vehicle, the vehicles
communicate with each other and aim to maintain a safe distance with each other,
under an occasional loss of communication. The Building is a Los Angeles hospital
building model with eight floors, each having three degrees of freedom. This is a
48-dimensional HA model. The Drivetrain models the engine of a vehicle together
with the other rotating components such as the gears, clutches, and differentials. The
backlash between the components is modeled via discrete transitions in the HA. For
each benchmark, the HA model parameters such as the number of locations and the
dimension are mentioned in the experiment tables below.
Experiment Setup:
The experiments of Table 1 are on Intel i7-4770, 3.40 GHz, 8 cores (hyper-threading
enabled), 8 GB RAM. The experiments of Table 2 are on a VM with 4 cores, 4 GB
RAM. The trajectories are computed with a time-step of 0.01. The simulation time
is computed for a fixed time-horizon or until reaching a fixed bound on the discrete
transitions.
Performance Evaluation:
Table 1 shows the performance comparison of computing 1000 random simulations
in the parallel simulation engine with the sequential counterpart and SpaceEx. In
the table, Nav(1), Nav(2), and Nav(3) denotes different instances of the Navigation
benchmark. We observe a maximum speedup of 5× and 76× in Building and Nav
(2) in the parallel-engine (XSpeed Par) over sequential (XSpeed Seq) simulation
and SpaceEx respectively. Table 2 shows a similar comparison over CORA. The TH
column shows the considered time-horizon of the computed simulations. A maximum
of 52× speedup is observed over CORA in Bouncing-Ball model.
12 K. Das et al.

Table 2 Performance comparison of 1000 random simulations over CORA

Time (in secs) Speedup
Benchmarks Dim. #Locs TH XSpeed CORA
Thermostat 2 2 30 1.84 50.25 27.31
B. Ball 3 1 30 1.82 94.91 52.15
FiveDimSys 5 1 10 0.88 5.68 6.45
Platoon 10 2 30 1.04 16.83 16.18
Drivetrain 10 4 10 3.51 135.29 38.54
Helicopter 29 1 20 6.79 23.17 3.41
Building 48 1 5 21.58 23.90 1.11

(a): Nav (3) (b): Thermostat (c): Helicopter (d): Drivetrain

Fig. 3 Computed simulation trajectories in XSpeed for various benchmarks

4.1 Correctness Evaluation

To evaluate the semantic correctness of our simulator, we compare the generated

trajectories with that of SpaceEx and CORA and found them to be matching. Some
of the computed trajectories on selected variables from four benchmarks are reported
in Fig. 3. During the evaluation, we detected few bugs in the SpaceEx simulator. For
example, Fig. 4a shows the trajectory in SpaceEx for Nav (2) that did not complete
till the time-horizon since it failed to detect the satisfaction of the jump condition
(x1 = 1 & x2 ∈ [3, 4]) for a simulation time-step of 0.05. For the same time-step,
our simulator in XSpeed correctly computes the entire trajectory shown in Fig. 4b.
In addition, the trajectory in SpaceEx for Nav (3) splits as shown in Fig. 4a. The
expected trajectory is computed by our simulator in XSpeed as shown in Fig. 5b.
The Green region in the figures denotes the vectors satisfying the Init predicate.

5 Conclusion

We present a parallel simulation engine for hybrid automaton models of CPS which
can compute random simulations in parallel on multicore processors. Our simula-
tion engine implements a multi-threaded lock-free algorithm in order to efficiently
Parallel Simulation of Cyber-Physical-Systems 13

Fig. 4 Premature termination of trajectory computation in SpaceEx

Fig. 5 SpaceEx trajectory splits for a single start state in Nav (3)

compute the simulations. We demonstrate performance speedup over SpaceEx and

CORA on standard CPS benchmarks.

Acknowledgements Rajarshi Ray gratefully acknowledges financial support from the Science and
Engineering Research Board (SERB) project with file number IMP/2018/000523. Amit Gurung is
grateful to Martin Luther Christian University, Shillong, Meghalaya, for partially supporting the
work under project grant No. Seed-Grant/559/2017-5567.
14 K. Das et al.

References

1. Althoff M (2010) Reachability analysis and its application to the safety assessment of
autonomous cars. PhD thesis, Technische Universität München
2. Althoff M, Grebenyuk D (2016) Implementation of interval arithmetic in cora 2016. In:
ARCH@ CPSWeek, pp 91–105
3. Alur, R.: Principles of cyber-physical systems. MIT Press (2015)
4. Alur R, Courcoubetis C, Henzinger TA, Ho PH (1992) Hybrid automata: an algorithmic
approach to the specification and verification of hybrid systems. In: Hybrid systems. Springer,
pp 209–229
5. Alur R, Dill DL (1994) A theory of timed automata. Theoretical computer science 126(2):183–
235
6. Antoulas AC, Sorensen DC, Gugercin S (2001) A survey of model reduction methods for
large-scale systems. Contemporary Mathematics 280:193–219
7. Bak S, Duggirala PS (2017) Hylaa: a tool for computing simulation-equivalent reachability
for linear systems. In: Proceedings of the 20th international conference on hybrid systems:
computation and control. ACM, pp. 173–178
8. Brand D, Zafiropulo P (1983) On communicating finite-state machines. Journal of the ACM
(JACM) 30(2):323–342
9. Coddington EA, Levinson N (1955) Theory of ordinary differential equations. Tata McGraw-
Hill Education
10. Damm W, Harel D (2001) Lscs: Breathing life into message sequence charts. Formal methods
in system design 19(1):45–80
11. Donze A (2010) Breach: a toolbox for verification and parameter synthesis of hybrid systems.
In: In Computer-aided verification, pp 167–170
12. Duggirala PS, Mitra S, Viswanathan M, Potok M (2015) C2e2: a verification tool for stateflow
models. In: International conference on tools and algorithms for the construction and analysis
of systems. Springer, pp 68–82
13. Fehnker A, Ivancic F (2004) Benchmarks for hybrid systems verification. In: HSCC, vol 4.
Springer, pp 326–341
14. Frehse G, Le Guernic C, Donzé A, Cotton S, Ray R. Lebeltel O, Ripado R, Girard A, Dang
T, Maler O (2011) SpaceEx: scalable verification of hybrid systems. In: Proceedings of CAV.
LNCS, vol 6806. Springer, pp 379–395
15. Gurung A, Deka A, Bartocci E, Bogomolov S, Grosu R, Ray R (2016) Parallel reachability
analysis for hybrid systems. In: 2016 ACM/IEEE international conference on formal methods
and models for system design (MEMOCODE). IEEE, pp 12–22
16. Gurung A, Ray R, Bartocci E, Bogomolov S, Grosu R (2018) Parallel reachability analysis of
hybrid systems in XSpeed. Int J Softw Tools Technol Transf 1–23
17. Hainry E (2008) Reachability in linear dynamical systems. In: Conference on computability
in Europe. Springer, pp 241–250
18. Henzinger TA (2000) The theory of hybrid automata. In: Verification of digital and hybrid
systems. Springer, pp 265–292
19. Holzmann GJ (2012) Parallelizing the SPIN model checker. In: Proceedings of SPIN 2012.
LNCS, vol 7385. Springer, pp 155–171
20. Jensen JC, Chang DH, Lee EA (2011) A model-based design methodology for cyber-physical
systems. In: 2011 7th international wireless communications and mobile computing conference.
IEEE, pp 1666–1671
21. Jin X, Deshmukh JV, Kapinski J, Ueda K, Butts K (2014) Powertrain control verification bench-
mark. In: Proceedings of the 17th international conference on Hybrid systems: computation
and control. ACM, pp. 253–262
22. Lee EA, Seshia SA (2016) Introduction to embedded systems: a cyber-physical systems
approach. MIT Press
23. Lygeros J, Tomlin C, Sastry S (1999) Hybrid systems: modeling, analysis and control. preprint
Parallel Simulation of Cyber-Physical-Systems 15

24. Makhlouf IB, Kowalewski S (2014) Networked cooperative platoon of vehicles for testing
methods and verification tools. In: ARCH@ CPSWeek, pp 37–42
25. Mathworks: Model-Based Design (2020). https://www.mathworks.com/solutions/model-
based-design.html
26. Paterno F (1999) Model-based design and evaluation of interactive applications. Springer Sci-
ence & Business Media
27. Ray R, Gurung A, Das B, Bartocci E, Bogomolov S, Grosu R (2015) Xspeed: accelerating
reachability analysis on multi-core processors. In: Piterman N (ed) Hardware and software:
verification and testing—11th international haifa verification conference, HVC 2015, Haifa,
Israel, November 17-19, 2015, Proceedings. Lecture Notes in Computer Science, vol. 9434.
Springer, pp 3–18. https://doi.org/10.1007/978-3-319-26287-1_1
28. Reisig W (2012) Petri nets: an introduction, vol 4. Springer Science & Business Media
29. Serban R, Hindmarsh AC (2005) Cvodes: the sensitivity-enabled ode solver in sundials. In:
ASME 2005 international design engineering technical conferences and computers and infor-
mation in engineering conference. American Society of Mechanical Engineers, pp 257–269
30. Skogestad S, Postlethwaite I (2005) Multivariable Feedback Control: Analysis and Design.
John Wiley & Sons
Attack Detection Scheme Using Deep
Learning Approach for IoT

Vikash Kumar, Sidra Kalam, Ayan Kumar Das, and Ditipriya Sinha

Abstract In recent years, the major concern of Internet of Things is cybersecurity

as attacks are increasing at an alarming rate with the rapid increase in connectivity
and integrity of various devices. It is necessary to detect the attack on the IoT devices
in real time as undetected attack on IoT devices for a longer period of time can result
in system unavailability to the end-users. In this paper, fog layer is introduced to
manage the resource constraint devices. The proposed Intrusion Detection System
is implemented in the fog layer. Deep learning is used to train the proposed system
so that it can classify the network traffic as attack and benign. The proposed scheme
is evaluated using the UNSW-NB15 and Bot-IoT dataset, which confirms that the
proposed model is able to classify the network traffic as attack and benign effectively.

Keywords Deep learning · Intrusion detection system · IoT · Security · Dataset

1 Introduction

Internet of Things (IoT) is an emerging technology that enabled the communica-

tion and processing of data in smart applications such as smart city, e-health, smart
agriculture, etc. The increasing application of IoT results in a huge increment in the
number of unknown cyber-attacks. Various sensitive and confidential data of users
are prone to numerous types of attacks from both internal and external attackers.
These types of attacks can be generated by either human or machine. For example,
the data breach of bitcoin leads to the loss of almost $70M and data breach of yahoo

V. Kumar · D. Sinha
National Institute of Technology Patna, Patna 800005, India
e-mail: [email protected]
D. Sinha
e-mail: [email protected]
S. Kalam · A. K. Das (B)
Birla Institute of Technology, Mesra, Patna campus, Patna 800015, India
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2021 17
R. Chaki et al. (eds.), Advanced Computing and Systems for Security: Volume 14,
Lecture Notes in Networks and Systems 242,
https://doi.org/10.1007/978-981-16-4294-4_2
18 V. Kumar et al.

caused the loss of $350M [1]. Traditional security measures cannot be applied to
IoT as they are low power and resource constrained. These issues can be resolved by
implementing fog layer, which is the extension of the cloud computing that enables
computing service to reside at the edge of the network. In fog layer, the deployed IoT
devices are of high capacity in terms of computational power and energy resource.
Thus, the task of Intrusion Detection System can be easily done in this layer, whereas
the low capacity devices of data sensing layer are deployed to sense different events.
This reduces the burden of the IoT as the storage, pre-processing and computation are
shifted to the nearby fog nodes. The security issue of the IoT generates the need for a
reliable Intrusion Detection System (IDS). An IDS is an application that detects the
malicious activity and classifies the data as malicious and benign at host level and at
network level. Network Intrusion Detection System (NIDS) uses network behavior to
detect attack and Host-based Intrusion Detection System (HIDS) uses system activ-
ities for the detection of attack. The proposed scheme will focus on the former one.
Network traffic is analyzed using anomaly detection and misuse detection. Misuse
detection matches the incoming traffic with already stored signatures for detecting
the attack. Database needs to be updated regularly for new types of attack. It cannot
detect an unknown attack on its own. Anomaly detection detects unknown attack
using behavioral analysis. In the proposed scheme, NIDS uses deep learning for the
training and classification of the attack and benign. The working of deep learning
is inspired by the way human brain thinks and takes the decision. It is an advanced
version of machine learning that is comprised of multiple layers. These layers are
used for feature extraction from the raw data. Each layer is trained to transform the
raw data into more intellectual and composite representation. The main motives of
this research are:
• Develop an anomaly-based intrusion detection model using deep learning
approach.
• Evaluation of the model for checking its efficiency.
The remainder of the paper is divided as—Sect. 2 deals with the study of related
field, Sect. 3 explains the overview of deep learning, Sect. 4 describes the proposed
work, Sect. 5 evaluates the performance of the proposed scheme, and the paper is
concluded in Sect. 6 followed by references.

2 Literature Survey

As IoT is nowadays a buzzword for the entire world, there come many barriers
along with it. Security is a major concern as it makes the system vulnerable to
many cyber-attacks. In order to resolve this issue, deep learning is used. Deep
learning has emerged as an advanced technique of machine learning. The tradi-
tional machine learning algorithms are less capable of attack detection as compar-
ative to the deep learning algorithm [2]. Multiple layer deep learning comprises of
multiple hidden layers, which encourages the model to detect the attacks in the IoT
Attack Detection Scheme … 19

network. There are many research works that discuss about the intrusion detection
using deep neural network. Nathan Shone and Tran Nguyen Ngoc use the non-
symmetric deep auto encoder (NDAE) for the intrusion detection [3]. They stacked
NDAEs in order to form deep learning hierarchy to deal with the complicated rela-
tionships between the features. NDAEs are comprised of multiple hidden layers,
which are non-symmetrical to learn the features from unlabeled data. Classification
of network traffic as normal or benign can be easily done by the neural network
concept. One more approach regarding this is Recurrent Neural Network (RNN)
that can also be used for malware detection [4]. Now a day, many new attacks are
attempting to exploit the system. In an attempt to protect the system against the
attack, first, we need to detect the attacks then only it can be prevented. Many deep
learning-based intrusion detection systems are developed [5] for this implementa-
tion. The attacker keeps on changing their methods every time, IoT network needs
an IDS, which should be flexible to deal with these issues. An intelligent intrusion
detection system is developed to detect and classify these unpredictable attacks [6]
in which network-based intrusion detection system (NIDS) and Host-based intrusion
detection system (HIDS) are combined to detect the cyber-attacks more efficiently.
When attacks are not detected for a longer period of time, it will affect the availability
of the system for the end user. It is very important to deal with these attacks in real
time to minimize the loss. An anomaly based intrusion detection system is devel-
oped, which implemented deep learning [7] to deal with these issues. This approach
worked successfully against wormhole attack, black hole attack, sinkhole attack,
DDoS and opportunistic service attack. DDoS attack is mostly common as it can be
easily spread on a larger scale. Restricted Boltzmann machines (RBM) specifically
handle the DDoS attack [8]. RBMs have the ability to learn the complex features in
an unsupervised learning system. In case of supervised learning, deep convolutional
neural network (DCNN) is also used. Hyun Min Song and Jiyoung Woo have also
developed a model, which uses DCNN in the IDS to provide security in Controller
Area Network (CAN) [9]. CAN is basically used to broadcast the information of the
current status of the vehicle. There are many more ways in which malicious activities
can govern the IoT system. It becomes very important to secure it from the various
types of cyber-attacks as it may lead to risk of life. Multiple IDS frameworks based
on deep learning are already in trend as it is successfully detecting the attacks and
simultaneously preventing the system with higher accuracy rate. Table 1 describes
the attacks detected in the existing schemes.

3 Deep Learning Overview

Deep learning is the breakthrough of the machine learning with increased accuracy
as compared with traditional learning algorithms. It is used for feature extraction and
training of the system. It is comprised of multiple consecutive layers that are used to
perform various operations. Each layer is interconnected to one another and output of
the previous layer is fed as an input to the next consecutive layer. Various application
20 V. Kumar et al.

Table 1 Attacks detected in the existing schemes

Authors Dataset Name of attack detected
Abebe et al. [10] NSL-KDD Probe, Remote to Local (R2L),
User to Root (U2R), Denial of
Service (DoS)
Nathan et al. [2] KDD CUP’99 and NSL-KDD Probe, R2L, U2R, DoS
Thamilarasu et al. [7] Dataset generated from smart home DDos, opportunistic, blackhole,
network wormhole and sinkhole
Moustafa et al. [11] NSL-KDD and UNSW-NB15 Probe, normal, Dos, R2L, exploits,
U2R, fuzzer, backdoor, generic,
normal, reconnaissance, shellcode,
and worms
Feng et al. [12] KDD CUP’99 DoS, cross site scripting (XSS)
and SQL

areas of deep learning are vehicle automation, natural language processing, image
processing, medical customer relationship management automation, etc. The most
common form of deep learning algorithm is multi-layer deep network. Activation
function computes the output of the previous layer and bias for the formation of
weighted input for the next layer. In deep learning algorithm, the value of output is
usually is set to either equal to input or less than the input. During feature extraction
in deep model, loss function is optimized to improve the efficiency of the model as
formulated in Eq. (1), which calculates the average of squared difference between
the predicted and actual resultant.
n  2
i=1 χ − χ̂i
Loss = (1)
n
The cumulative loss function for a set of n number of training data is defined in
Eq. (2).

1 n  2 λ l nl nl+1 (l)2

L(w) = χi − χ̂i + Wmk (2)
2n i=0 2 j=1 k=1 m=1

where, l denoted the number of layers and nl represents the number of nodes in each
layer.
Stochastic Gradient Descent method is used to minimize the loss. Error minimiza-
tion is done using weight and activation function as in Eq. (3) where ωij represents
the updated weight of the link connecting ith and jth neuron of two consecutive layers
δε
and δω ij
represents the error gradient with respect to the weight ωij .

δε
ωi j = η (3)
δωi j
Attack Detection Scheme … 21

4 Proposed Work

Fog layer is responsible for the communication between the users and sensor. Attack
detection system is deployed in the fog layer as they are at the edge of the cloud layer
and close to the interacting users. The proposed intrusion detection model is depicted
in Fig. 1. Deep neural network is used in IDS for attack detection, which is better
than other traditional machine learning algorithm. DNN is best suited for resource-
constrained low power IoT devices because of their thin and layered structure. Three
phases of the intrusion detection system that are used in the proposed scheme are
feature extraction, training and decision-making, which is described in the below
subsection.

4.1 Feature Set and Data Pre-processing

IDS categorizes the malicious and benign communication on the basis of behavior
of the message. A single message cannot determine the behavior of the message.
Feature of communication cycle needs to be observed for a period of time for deter-
mining the behavior of the message. In a normal communication between a pair
of sender and receiver, rate of message sent and rate of message received must be
similar. If there is a difference between the rate of transmission and reception then
there must be some attack taking place like Denial of service attack, Replay attack,
sinkhole attack, etc. Based on the training, IDS classifies the message as malicious
and benign message. IDS classifies the message as malicious and benign on the
basis of reception rate, transmission rate, IP of source, IP of destination, transmis-
sion mode, duration, information of data value, etc. All these features are selected

Fig. 1 Proposed intrusion-detection model

22 V. Kumar et al.

Fig. 2 Overview of the proposed DNN training

according to the computational capability of the low-powered resource constrained

IoT devices. In the data pre-processing phase, label encoding is used to translate the
data into machine readable format and feature scaling is used for the standardization
of the distinct features that are available in the data into a fixed range. The features are
feed to the deep neural learning model for the training purpose so that it can classify
the traffic as malicious and benign. The working model of the proposed scheme is
depicted in Fig. 2.

4.2 Training

The proposed model consists of five-layer deep learning model, which consists of
one input layer, three hidden layers and one output layer, which classifies the network
traffic as malicious and benign. The first layer consists of 1024 neurons and hidden
layers consist of 768, 512 and 256 neurons, respectively. Feature vector f v is fed
to the Deep Neural Network (DNN) in the first layer and it passes through all the
Attack Detection Scheme … 23

layers. Each hidden layer is connected to the next hidden layer and provides the
filtered output to the next layer using ReLu activation function as defined in Eq. (4).
The output is calculated by the output layer and it uses Softmax activation function
for the generation of output, which is defined in Eq. (5).

x = max(0, z) (4)

∈χi
ρ(χi ) = κ (5)
j=1 ∈χ j

In the training phase, feature vector is fed through the external node to DNN,
which is present at the bottom of the DNN. The weight attached to each neural node
is initialized. As data pass from the DNN layers in each cycle of training, weight
needs to be modified consequently.

4.3 Decision-Making

Once the training phase is done, the system makes the decision regarding the catego-
rization of the network traffic. It, this phase network traffic, is classified as malicious
and benign on the basis of the training provided to the DNN layer.

5 Experimental Results

In this section, the proposed scheme is validated and evaluated. Keras is used for the
implementation of DNN, which is an open source neural network library [4].

5.1 Dataset

KDDCUP99, NSL-KDD, CICIDS, Bot IoT and UNSW-NB15 are the widely used
datasets for the research work of intrusion detection. The proposed scheme has used
the UNSW-NB15 and Bot-IoT dataset, which is available in .csv format for the
purpose of evaluation of the model. UNSW-NB15 has resolved the issues found in
KDDCUP99 and NSL-KDD. UNSW-NB15 dataset is available in two forms. One
consists of 2 million records and the second is a partition of full record dataset, which
consists of 42 features and class is categorized as normal and nine types of attacks.
Detailed analysis of the dataset is described in Table 2.
24 V. Kumar et al.

Table 2 Testing record distribution of UNSW_NB15

Class Description Number of records
Normal Legitimate user data 37,000
Fuzzers It attempts to suspend the network or program by feeding 6,062
the randomly generated data. It consists of the attacks
related to html file penetration, spams, etc.
Analysis It contains attacks related to spam, port scan and html file 677
penetration
Backdoor It is a technique in which security of the system is evaded 583
to access the system
DoS Attacker makes the network resource unavailable to the 4,089
legitimate user by interrupting the services
Exploits In this, the attacker is aware of the loophole of the 11,132
operating system or the application software and its aim
is to exploit the vulnerability
Generic It works against all block cipher irrespective about the 18,871
structure of the block cipher
Reconnaissance An attacker observes the target system for gathering 3,496
information for vulnerability
Shellcode It is a small piece of the program used as the payload for 378
the exploitation of software vulnerability
Worms It replicated itself and spread to other computer using 44
computer networks. It relies on security failures of the
target system

Bot IoT dataset was created by deploying a real network environment, which
consists of botnet and normal traffic. The dataset consists of various attack details,
which are discussed in Table 3.

Table 3 Testing record distribution of Bot-IoT dataset

Class Description Number of records
Normal Legitimate user data 107
DoS Attacker makes the network resource unavailable to the 330,112
legitimate user by interrupting the services
DDoS Distributed Denial of service makes the network 385,309
resource unavailable by using multiple sources
Reconnaissance Attacker observes the system to gather the information 18,163
for vulnerability
Theft It refers to the stealing of confidential data or 14
information
Attack Detection Scheme … 25

5.2 Results Analysis

The proposed scheme is evaluated using the evaluation metrics such as accuracy,
precision, recall and F-1 score where TP represents true positive, TN represents true
negative, FP represents false positive and FN represents false negative.
Accuracy is defined as the total number of correct prediction of data to the
total number of data instances. It is one of the important factors for evaluating the
performance of any IDS, which is defined in Eq. (6).

TP + TN
Accuracy = (6)
TP + TN + FP + FN

The ratio of correctly predicted positive data and total number of data in the
actual class is called Recall. It basically indicates the number of attacks returned by
the system, which is defined in Eq. (7).

TP
Recall = (7)
TP + FN

Precision is defined as the ratio of correctly predicted positive data and total
predicted positive data. It predicts the number of attacks returned correctly, which is
defined in Eq. (8).

TP
Precision = (8)
TP + FP

The harmonic mean of recall and precision is called F1 score, which reflects the
stability between both of them, which is defined in Eq. (9).

2 ∗ (precision ∗ recall)
F1 score = (9)
precision + recall

Table 4 describes the performance of the system for two classes, namely, Normal
and Attack on UNSW-NB15 dataset for six hidden layers, whereas Table 5 describes
the performance for nine classes on the same dataset for three hidden layers. Table
6 describes the performance of the system for five classes on Bot-IoT dataset and
Table 7 describes the performance for seven class attack subcategories on the same
dataset for three hidden layers. Finally, Fig. 3 compares the accuracy of the proposed
system for both the dataset.
26 V. Kumar et al.

Table 4 Performance of 2-class on UNSW-NB15

Architecture Neurons Class Precision (%) Recall (%) F1 score (%)
Hidden layer 0 0 Normal 97 58 73
Attack 74 99 85
Hidden layer 1 768 Normal 97 67 79
Attack 79 98 87
Hidden layer 2 512 Normal 97 93 95
Attack 94 98 96
Hidden layer 3 256 Normal 97 84 90
Attack 88 98 93
Hidden layer 4 128 Normal 94 58 72
Attack 74 97 84
Hidden layer 5 64 Normal 97 49 65
Attack 70 99 82
Hidden layer 6 32 Normal 100 89 94
Attack 92 100 96

6 Conclusion

In this paper, a network intrusion detection system is proposed for fog assisted IoT
system. The proposed model is implemented using python and extensive evaluation
is performed. The benchmark UNSW-NB15 and Bot IoT dataset is used for the eval-
uation of the model. The experiment shows that DNN is implemented successfully
for the attack detection in fog-assisted IoT. The model is evaluated on the basis of
precision, recall and F1 score and obtained promising results. To evaluate the perfor-
mance, we have followed two test cases: classification of records as normal and attack
with all the features and classified the attack into its categories with all the features.
Attack Detection Scheme … 27

Table 5 Performance of 9-class on UNSW-NB15

Architecture Class Precision (%) Recall (%) F1 score
Hidden layer 1 Normal 100 100 100
Backdoor 0 0 0
Analysis 0 0 0
Fuzzers 62 61 61
Shellcode 0 0 0
Reconnaissance 44 46 45
Exploits 57 91 70
DoS 0 0 0
Worms 0 0 0
Generic 100 95 98
Average score: 85.41
Hidden layer 2 Normal 100 100 100
Backdoor 0 0 0
Analysis 0 0 0
Fuzzers 87 48 62
Shellcode 0 0 0
Reconnaissance 47 72 57
Exploits 53 94 67
DoS 52 87 65
Worms 0 0 0
Generic 100 89 94
Accuracy score: 80.28
Hidden layer 3 Normal 100 72 84
Backdoor 0 0 0
Analysis 0 0 0
Fuzzers 83 56 67
Shellcode 0 0 0
Reconnaissance 47 58 52
Exploits 45 94 61
DoS 0 0 0
Worms 0 0 0
Generic 100 100 100
Accuracy score: 78.72
28 V. Kumar et al.

Table 6 Performance of 5-class on Bot IoT

Architecture Class Precision Recall F1 score
Hidden layer 1 Normal 100 100 100
Reconnaissance 81 100 90
DDoS 100 100 100
DoS 100 100 100
Theft 0 0 0
Accuracy score: 99.98
Hidden layer 2 Normal 100 100 100
Reconnaissance 88 100 94
DDoS 100 100 100
DoS 100 100 100
Theft 33 7 12
Accuracy score: 99.99
Hidden layer 3 Normal 100 100 100
Reconnaissance 99 100 100
DDoS 100 100 100
DoS 100 100 100
Theft 27 100 43
Accuracy score: 99.99
Attack Detection Scheme … 29

Table 7 Performance of 7-class of attack subcategories on Bot IoT

Architecture Class Precision Recall F1 score
Hidden layer 1 UDP 100 98 99
TCP 100 100 100
Service_Scan 28 21 24
OS_Fingerprint 11 26 15
HTTP 4 53 8
Normal 2 34 3
Keylogging 0 0 0
Accuracy score: 96.70
Hidden layer 2 UDP 100 21 35
TCP 58 100 73
Service_Scan 98 42 59
OS_Fingerprint 29 96 44
HTTP 1 94 1
Normal 0 0 0
Keylogging 0 0 0
Accuracy score: 97.39
Hidden layer 3 UDP 100 100 100
TCP 100 100 100
Service_Scan 98 13 22
OS_Fingerprint 22 93 35
HTTP 39 96 55
Normal 22 88 36
Keylogging 0 0 0
Accuracy score: 98.08

100

80 UNSW NB-15
Accuracy

60
Bot IoT
40

20

0
1 2 3
Number of Hidden Layers

Fig. 3 Accuracy of UNSW NB-15 and Bot IoT

Exploring the Variety of Random
Documents with Different Content
possessed neither the vigour nor the prudence which circumstances
demanded.
The rumours which soon began to circulate at Rome of the perils
which Cæsar was incurring at Alexandria, rendered his conduct
uncertain; he hesitated to put down, with a firm hand, the disturbers
of the republic, whom the death of his master might make more
powerful than himself. The son-in-law of Cicero, Cornelius Dolabella,
overwhelmed with debt, had followed the example of Clodius in
getting himself adopted by a plebeian, and had thus acquired the
tribunate. In this position he had recommended himself, like Cælius,
to the worst classes of the citizens, by urging an abolition of debts.
One of his colleagues resisted, and both betook themselves to
violence. For some time Antony looked on as if uncertain which party
to espouse; but a domestic affront from Dolabella, who had
intrigued with his wife, roused his passion; he attacked the turbulent
mob with arms, and filled the streets with the indiscriminate
slaughter of eight hundred citizens. He did not venture, however, to
punish the author of the disturbance, but contented himself with
menaces and precautions till the fortunate arrival of the dictator
himself in September, 47.

Roman Battering-ram with Testudo

Contrary to the apprehensions of many of the citizens Cæsar’s

return was marked by no proscription. He confined himself to the
confiscation of the estates of the men who still remained in arms
against him; and that of Pompey himself, whose sons were in the
hostile camp, he set up to public auction. A portion of them was
bought by Antony, who ventured to evade the due payment of the
price. He conceived that his services might command the trifling
indulgence of release from a paltry debt. He found, however, that his
patron was in earnest, and prudently submitted to the affront. The
dictator remained only three months in Rome. Every moment was
fully occupied in the vast work of reconstructing the government;
but we know not what were the special measures enacted at this
period, and Cæsar’s legislation may fitly be reserved to be
contemplated hereafter at a single view. Two consuls were
appointed for the remaining three months of the year, and for the
next ensuing Cæsar nominated himself for the third time, together
with Lepidus. He caused himself also to be again created dictator.
His partisans he loaded with places and honours, and sated the
populace with largesses. The soldiers demanded the fulfilment of his
repeated promises. Those of the tenth legion broke out into open
revolt, and ran from Campania to Rome to extort their claims. Cæsar
convoked them in the Field of Mars, approached them unattended,
mounted his tribunal, and demanded the statement of their
grievance. At the sight of their redoubted general their voices
faltered, their murmurs died away; they could only ask for their
discharge. “I discharge you, citizens,” replied the imperator; and
they cowered under this disparaging appellation, abashed and
humiliated. To the fierce and haughty soldier the peaceful name of
citizen seemed a degradation. They entreated to be restored to their
ensigns, and submitted to severe punishment in expiation of their
fault. This simple incident is a key to the history of the times. This
application of the title of citizen, and the effect it produced, show
plainly that the basis of Cæsar’s force was purely military, and that
Cæsar himself knew it. This was the point at which every party
leader in turn had tried for years to arrive, and Cæsar had
succeeded.

THE AFRICAN WAR

As soon as this sedition was repressed Cæsar departed to crush

the remnant of his enemies assembled in Africa. The defeated host
had been scattered in various directions, but the largest division of
the fugitives had made its way to Dyrrhachium, and there
deliberated on its further movements. Cato, to whom the command
was offered, waived it in favour of Cicero, as his superior in rank; but
the orator declined to associate himself further in the honours and
perils of a fruitless struggle, and departed mournfully for Italy. His
life was with difficulty preserved from the fury of Cneius, the elder
son of the great Pompey, a man of ungovernable passions and
slender capacity. Shortly afterwards Scipio assumed the command of
the main body, and carried it to Utica in the province of Africa. Cato
at the head of another division skirted the coasts of Greece and Asia,
and picked up some scattered adherents of the cause. He followed
in the track of Pompey, but when the news of his chief’s
assassination reached him, he landed on the shore of Libya, and
demanded admission within the walls of Cyrene. The natives shut
their gates; but Cato, always loath to exercise any unprofitable
severity, generously abstained from chastising them. Anxious now to
effect a junction with the remainder of his friends, he coasted
westward as far as the lesser Syrtis, and then plunged with his little
army into the sandy desert. The seven days’ march through this
inhospitable region, torrid with heat and infested with serpents, was
justly considered one of the noblest exploits of the Roman
legionaries. The poet of the Pharsalia exalts it above the three
triumphs of Pompey and the victories of Marius over the tyrant of
Numidia. He turns with pardonable enthusiasm from the deified
monsters, the Caligulas and Neros of his own day, to hail its achiever
as the true Father of his Country, the only worthy object of a free
man’s idolatry.
The arrival of Cato at the headquarters of the republicans in Utica
was quickly followed by that of Cneius Pompey, and in the course of
the year 47 the remains of the great host of Pharsalia were
assembled with many reinforcements under the banners of Scipio.
These forces amounted to not less than ten complete legions, and
Juba, who could bring one hundred and twenty elephants into the
field, besides innumerable squadrons of light cavalry, had promised
his assistance. The officers began to brag of their future triumphs
almost as loudly as before their recent disasters. Their defiance was
re-echoed to the opposite shores of Italy, and caused fresh dismay
to the time-servers, who had abandoned the Pompeian cause on the
event of its first discomfiture. But this force, numerous as it was,
was not in a condition, it would seem, to choose a distant field of
operations. The want of money may have compelled its chief still to
act on the defensive, and await through a whole year the expected
attack of the enemy. Nor were these chiefs themselves unaffected by
personal jealousies. Scipio and Varus contended for the command,
the one as the foremost in rank and dignity, the other as the
legitimate proconsul of the province; while Juba, conscious of his
own importance to the cause, affected to lord it over both. Cato
alone continued still to act with his usual simplicity of purpose and
patriotic devotion. But his noble demeanour rebuked the selfishness
of his associates, and they contrived to remove him from their
counsels by charging him with the defence of Utica, while they
shifted their own quarters to the neighbourhood of Hadrumetum.
The brave philosopher rejoiced that he was not compelled to draw
his sword in civil strife, while he busied himself not the less earnestly
in the collection of stores and preparation of defence. Of all the
professed asserters of Roman liberty he alone really lamented the
necessity of arming in her cause; from the first outbreak of the war
he had refused to trim his venerable locks or shave his grizzled
beard, and from the fatal day of Pharsalia he had persisted in sitting
at his frugal meals, and denied himself the indulgence of a couch.
A whole year had now passed, while the
[47-46 b.c.] republicans contemplated with folded arms the
perils Cæsar had surmounted in Alexandria, the
victory he had gained over Pharnaces, and the brilliant reception he
had met with in Rome. Cæsar assembled six legions and two
thousand horse at Lilybæum in Sicily, and in the middle of October
47, he appeared off the African coast with the first division of his
forces, and summoned the republicans in their camp at Hadrumetum
to surrender to “Cæsar the imperator.” “There is no imperator here
but Scipio,” they replied, and inflicted death upon his envoy as a
deserter. The dictator sailed on to Leptis, and was there invited to
take shelter, while he awaited the arrival of the rest of his
armament.
While these reinforcements were coming slowly in he was
attacked by Scipio, and subjected to annoyance and peril from the
movements of the enemy’s cavalry. Labienus, who frequently
charged him at the head of the Roman horse, distinguished himself
by the bitter taunts with which he addressed the veterans whom he
had so often led to victory. But Cæsar maintained himself in a
fortified position till he could move forward with a force of five
legions. At the same time the alliance he had formed with the
Mauretanian kings, Bogudes and Bocchus, the jealous rivals of the
Numidians, enabled him to draw off Juba to the defence of his own
capital Cirta. He pushed on, offering battle, which Scipio, though
with double his numbers, steadily refused, until Juba returned with
his vaunted elephants and cavalry. The necessities of the Roman
chiefs compelled them to submit to revolting indignities at the hands
of this barbarian ally. He forbade Scipio the use of the imperator’s
purple cloak, which he declared to belong only to kings. When he
issued his royal mandates to the Roman officers, they were observed
to be even more punctually obeyed than the orders of the general
himself.
At last on the 4th of April the armies met on the field of Thapsus.
On this occasion many of Cæsar’s men were fresh recruits, and he
was not without some misgivings about their steadiness. But they
were not less impatient for the onset than the veterans, whom their
general recommended to their imitation, and loudly demanded the
signal to engage. While he still hesitated, checking with hand and
voice the impatient swaying of the lines, suddenly the blast of a
single trumpet burst forth on the right wing. The impetuous ferocity
of the tenth legion could no longer brook restraint; they had raised
the signal unbidden; and now the whole army rushed forward in one
unbroken body, overpowering their officers’ efforts to detain them.
Cæsar, when he beheld rank after rank pouring by him, without the
possibility of recall, gave the word “Good luck” to his attendants, and
spurred his horse to the head of his battalions. The combat was
speedily decided. The elephants, thrown into confusion by the first
discharge of stones and arrows, turned upon the ranks they were
placed to cover, and broke in pieces their array. The native cavalry,
dismayed at losing their accustomed support, were the first to
abandon the field. Scipio’s legions made little resistance; they sought
shelter behind their entrenchments. But their officers had fled, and
the men, left without a commander, rushed in quest of their
discomfited allies. They found the Numidian camp in the hands of
the enemy; they begged for quarter, but little mercy was shown
them, and Cæsar himself beheld with horror a frightful massacre
which he was powerless to control. Scipio escaped to the coast, and
embarked with others for Spain, but was intercepted and slain.[119]
Juba and Petreius fled together, and sought refuge within the walls
of Zama. But the Numidians rejoiced in the defeat of their tyrants
and refused them solace or shelter. The fugitives, repulsed in every
quarter, and disdaining to solicit the victor’s clemency, placed
themselves at a banquet together, drank their fill of wine, and
challenged each other to mortal combat. Petreius, the elder of the
two, was despatched by his opponent, who then threw himself upon
his own sword.[120]
The rout of Thapsus was known at Utica on the same evening. On
the morrow Cato convened the Roman officers and residents, and
laid before them the state of their affairs. Calmly and cheerfully he
enumerated his means of defence, and desired them to decide for
themselves whether they would resist the conqueror, or seek safety
in flight or capitulation. The knights and senators, despairing of
pardon, would have held out to the uttermost; but the traders and
men of peace, who had long settled in Utica, and were conscious
that they had done nothing hitherto to provoke the wrath of the
assailant, insisted on a timely surrender. When it was known that
Cæsar was approaching, Cato caused all the gates to be closed
except that which led to the sea, and urged all that would to betake
themselves to the ships. He dismissed his personal friends, of whom
a few only, and among them his own son, insisted on remaining with
him; for he had plainly intimated that for his own part he would not
quit his post. With these cherished associates he sat down to supper,
and discoursed with more than his usual fervour on the highest
themes of philosophy, especially on the famous paradox of the
stoics, that the good man alone is free, and all the bad are slaves.
His companions could not fail to guess the secret purpose over
which he was brooding. They betrayed their anxiety only by silent
gestures; but Cato, observing the depression of their spirits, strove
to reanimate them, and divert their thoughts by turning the
conversation to topics of present interest.
The embarkation was at this moment
[46 b.c.] proceeding, and Cato repeatedly inquired who
had already put out to sea, and what were the
prospects of the voyage. Retiring to his chamber he took up the
Dialogue on the Soul, in which Plato recorded his dying master’s last
aspirations after immortality. After reading for some time he looked
up and observed that his sword had been removed. In the irritation
of the moment he gave way to a burst of violence, such as often
marked the behaviour of the Roman master to his slave; calling his
attendant to his presence he struck him on the mouth, bruising his
own hand with the blow. He then sent for his son and friends, and
rebuked them sharply for their unworthy precaution; “as if,” he said,
“I needed a sword to kill myself, and might not, if I chose, put an
end to my existence by dashing my head against the wall, or merely
by holding my breath.” Reassured perhaps for the moment by the
calmness of his demeanour, they restored him his weapon, and at
his earnest desire once more left him alone. At midnight, still
anxious about those who were departing, he sent once again to
inquire if the embarkation were completed. The messenger returned
with the assurance that the last vessel was now on the point of
leaving the quay. Thereupon Cato threw himself on his bed, as if
about to take his rest for the night; but when all was quiet he seized
his sword and thrust it into his stomach. The wound was not
immediately mortal, and the victim rolled groaning on the floor. The
noise at once summoned his anxious attendants. A surgeon was at
hand, and the sufferer was unconscious while the protruding
intestines were replaced, and the gash sewn up. But on coming to
himself he repulsed his disconsolate friends, and tearing open the
fatal wound, expired with the same dogged resolution which had
distinguished every action of his life.

Death of Cato

(From a drawing by Mirys)

Cato had no cause to despair of retaining life under the new

tyranny. At an earlier period he had meditated, in such a
contingency, seeking refuge in retirement and philosophy. But his
views of the highest good had deepened and saddened with the fall
of the men and things he most admired. He now calmly persuaded
himself that with the loss of free action the end of his being had
failed of its accomplishment. He regarded his career as prematurely
closed, and deemed it his duty to extinguish an abortive existence.
[121] Cæsar, when he heard of his self-destruction, lamented that he
had been robbed of the pleasure of pardoning him, and to his
comrades in arms he exhibited, according to the most credible
accounts, the same clemency by which he had so long distinguished
himself. But the same man who could now speak and act thus
generously, did not scruple, at a later period, to reply to Cicero’s
panegyric with a book which he called the Anti-Cato, in which he
ridiculed the sage’s vain pretensions, and scoffed at him for raking in
his brother’s ashes for the golden ornaments of his pyre, for
transferring to Hortensius the wife who had borne him as many
children as he desired, and taking the widow to his arms again
enriched with a magnificent dowry. Could the proud philosopher
have anticipated a time when the wantonness of power might sport
unchecked with the good fame of its victims, he would have shrunk
from such moral degradation with greater horror than from the
servitude of the body.c

SALLUST’S COMPARISON OF CÆSAR AND CATO

“After hearing and reading of the many glorious achievements

which the Roman people had performed at home and in the field, by
sea as well as by land, I happened to be led to consider what had
been the great foundation of such illustrious deeds. I knew that the
Romans had frequently, with small bodies of men, encountered vast
armies of the enemy; I was aware that they had carried on wars
with limited forces against powerful sovereigns; that they had often
sustained, too, the violence of adverse fortune; yet that, while the
Greeks excelled them in eloquence, the Gauls surpassed them in
military glory. After much reflection, I felt convinced that the
eminent virtue of a few citizens had been the cause of all these
successes; and hence it had happened that poverty had triumphed
over riches, and a few over a multitude. And even in later times,
when the state had become corrupted by luxury and indolence, the
republic still supported itself, by its own strength, under the
misconduct of its generals and magistrates; when, as if the parent
stock were exhausted, there was certainly not produced at Rome, for
many years, a single citizen of eminent ability. Within my
recollection, however, there arose two men of remarkable powers,
though of very different character, Marcus Cato and Caius Cæsar,
whom, since the subject has brought them before me, it is not my
intention to pass in silence, but to describe, to the best of my ability,
the disposition and manners of each.
“Their birth, age, and eloquence, were nearly on an equality; their
greatness of mind similar, as was also their reputation, though
attained by different means. Cæsar grew eminent by generosity and
munificence; Cato by the integrity of his life. Cæsar was esteemed
for his humanity and benevolence; austereness had given dignity to
Cato. Cæsar acquired renown by giving, relieving, and pardoning;
Cato by bestowing nothing. In Cæsar there was a refuge for the
unfortunate; in Cato, destruction for the bad. In Cæsar, his easiness
of temper was admired; in Cato, his firmness. Cæsar, in fine, had
applied himself to a life of energy and activity; intent upon the
interests of his friends, he was neglectful of his own; he refused
nothing to others that was worthy of acceptance, while for himself
he desired great power, the command of an army, and a new war in
which his talents might be displayed. But Cato’s ambition was that of
temperance, discretion, and, above all, of austerity; he did not
contend in splendour with the rich, or in faction with the seditious,
but with the brave in fortitude, with the modest in simplicity, with
the temperate in abstinency; he was more desirous to be, than to
appear, virtuous; and thus, the less he courted popularity, the more
it pursued him.”e[122]

FOOTNOTES

[119] [Florusd says: “Scipio got off in a ship but, as the enemy
overtook him, he thrust his sword into his bowels; and when
some one asked where he was, he returned this answer: ‘The
general is well.’” Appianf says: “he ran his sword through his
body, and threw himself into the sea.”]
[120] [Says Florusd: “Petreius slew both Juba and himself; and
the half-consumed meats and funeral dishes were mixed with the
blood of a king and a Roman.”]
[121] [Florusd in Roman fashion says: “Hearing of the defeat of
his party, he did not hesitate to die; but even cheerfully, as
became a wise man, hastened his own death.”]
[122] [Sallust’s comparison of Cæsar and Cato should not
mislead the reader as to the importance of the latter, who in fact
exercised little influence on the great events of his age.]
 CHAPTER XXV. THE CLOSING SCENES OF
CÆSAR’S LIFE
THE END OF THE AFRICAN WAR

The suicide of Cato was the consistent act of a heathen

philosopher, determined at least to maintain the purity of his soul
uncontaminated by base compliances. Assuredly the calm dignity of
its execution demands our respect and compassion, if not the
principle on which it was based. Far different was the manner in
which the rude barbarian Juba and the coarse soldier Petreius ran
forward to meet their ends. They had escaped together from the
field of battle, and the Numidian offered to provide shelter for his
companion in one of his own strongholds. The Roman province was
so ill-disposed towards the barbarian chief that he was obliged to
hide himself by day in the most secluded villages, and roam the
country on his homeward flight during the hours of darkness. In this
way he reached Zama, his second capital, where his wives and
children, together with his most valuable treasures, were deposited.
This place he had taken pains to fortify at the commencement of the
war, with works of great extent and magnitude. But on his
appearance before the walls, the inhabitants deliberately shut their
gates against him and refused to admit the enemy of the victorious
Roman. Before setting out on his last expedition, Juba had
constructed an immense pyre in the centre of the city, declaring his
intention, if fortune went ill with him, of heaping upon it everything
he held most dear and precious, together with the murdered bodies
of the principal citizens, and then taking his own place on the
summit, and consuming the whole in one solemn conflagration. But
the Numidians had no sympathy with this demonstration of their
sovereign’s despair, and resolved not to admit him within their walls.
Juba having tried in vain every kind of menace and entreaty, to
which no reply was vouchsafed, at last retired, but only to
experience a similar reception in every other quarter to which he
resorted. He at least had little to hope from the clemency which the
victor had extended to his conquered countrymen. His companion,
hard as his own iron corslet, scorned to accept it. The fugitives
supped together, and, flushed with the fumes of the banquet,
challenged each other to mutual slaughter. They were but unequally
matched; the old veteran was soon despatched by his more active
antagonist, but Juba was constant in his resolution, and only
demanded the assistance of an attendant to give himself the last
fatal stroke.
Nor was the fate of Considius, of Afranius, and Faustus Sulla less
disastrous. The first of these had abandoned the defence of
Thysdrus at the approach of the forces which Cæsar despatched
against it, and attempted to make his escape with the treasures he
had amassed into the territories, until now friendly, of the Numidian
chieftains. He was destroyed, for the sake of his hoarded booty, by
the Gætulians who accompanied him in his flight. The others had
retained the command of a squadron of Scipio’s cavalry, and after
burning one town which had shut its gates against them had made a
desperate attack on the military post which Cato maintained outside
the walls of Utica, to wreak an unworthy vengeance on the Cæsarian
partisans there kept in custody. Baffled in this object they had made
their way into Utica, while Cato still commanded there, and had
added bitterness to his last days by the violence and ferocity of their
behaviour. From thence they led their ruffians along the coast in the
hope of finding means of transporting them into Spain. But on their
way they fell in with Sittius, who was advancing to join Cæsar; their
men were routed and themselves taken. The bands of the Roman
adventurer carried on war with the same brutality as the barbarians
among whom they practised it. The captors quarrelled among
themselves; their passions were inflamed, perhaps, in the
distribution of the prisoners and the booty; and both Afranius and
Faustus were killed in the fray which ensued. But the massacre of
the son of the dictator Sulla, accidental as it was, or at least
unauthorized, could hardly fail of being charged as a deliberate act
upon the representative of Marius.
While his foes were thus flying and falling, Cæsar advanced
triumphantly from the scene of his last exploit, receiving the
submission of the towns on his way, carrying off the stores and
treasure collected for his enemies’ use, and leaving garrisons to
retain them in fidelity. As he drew near to Utica he was met by L.
Cæsar, whose petition for mercy seems to have been confined to his
own person, and to whom, as well as to a long list of distinguished
nobles, the conqueror extended the promise of his protection. He
lamented with every appearance of sincerity that Cato had robbed
him of the pleasure of pardoning one who, of all his antagonists, had
been the most obstinate in his opposition, and the most inveterate in
his hatred. The fatal compliance of the Utican senators, who, not
content with obeying his enemies’ commands, had contributed
money to their cause, furnished him with a specious pretence for
rifling their coffers of the treasures he now most urgently needed.
His requisitions amounted to two hundred millions of sesterces. At
the same time the city of Thapsus was mulcted in two millions, and
the company of Roman traders in three. Hadrumetum paid down
three millions, and its Roman capitalists five. Leptis and Thysdrus
also suffered in due proportion. A grand auction was held at Zama
for the sale of all the objects of Juba’s royal state, and of the goods
of the Roman citizens who had borne arms under the tyrant’s orders.
Upon the people who had so boldly defied their sovereign, and
refused him admittance within their walls, honours and largesses
were munificently showered, and the taxes heretofore demanded for
the royal treasury were partially remitted by the collectors of the
republic. But the country of Numidia was deprived of its
independence, and definitely reduced to the form of a province,
under the proconsulate of Sallust. The rewarded and the punished
acquiesced equally in the conqueror’s dispositions; the submission of
Africa to his authority was from thenceforth complete. The Uticans
were allowed to commemorate with a funeral and a statue the
humane and noble conduct of their late governor.

THE RETURN TO ROME

Cæsar settled the affairs of Africa with his usual despatch, and
sailed from Utica on the fourteenth day of April, 46 b.c. On his way
to Italy, he stopped at Caralis, in Sardinia. The aid which the island
had afforded to his adversaries furnished him with a decent pretext
for extorting from the inhabitants large sums of money. At the end
of the same month he again weighed anchor; but the prevalence of
easterly winds drove him repeatedly to shore, and he at last reached
Rome on the twenty-eighth day after his departure from the
Sardinian capital. The reports he received at this time of the revival
of the republican cause in Spain did not give him much uneasiness.
Cneius had been detained by sickness in the Baleares, and the
fugitives from the field of Thapsus had been almost all cut off in
their attempts to reach the point to which their last hopes were
directed. The legionaries who had mutinied against Cassius Longinus
were still either unsatisfied with their treatment under the
commander who had superseded him, or fearful of their general’s
vengeance when a fitting opportunity should arrive. It was from
Cæsar’s own soldiers that the invitation had gone forth to the
republican chiefs to renew the struggle on the soil of Spain. The
spirit of the old commonwealth still survived in many of the towns of
Bætica; promises of support were freely given; but the remnant of
the African armament was contemptible both in numbers and ability.
Of all the haughty nobles who had thronged the tent of Pompey at
Luceria or Thessalonica, not one with a name known to history
remained in arms, except Labienus alone. He indeed had succeeded
in making his escape from Africa, in company with Varus; but the
insurgents had already placed themselves under the command of
Scapula and Aponius, officers of their own, nor would they suffer
themselves to be transferred from them to any other except the son
of the great Pompey. The extent to which the flame of insurrection
had spread was probably unknown at this time to Cæsar. He was
impatient to reap at last the fruit of so much bloodshed, to assume
the post of honour he had won, and to work out the principles and
objects of so many years of anticipation. A distant and contemptible
outbreak might be subdued without meeting it in person.
Accordingly, C. Didius, an officer of no eminent reputation, was sent
with a naval and military force to the succour of Trebonius, whom,
however, he found already expelled from his government by the
growing force of the new movement.
Meanwhile Rome had sunk, during the conqueror’s absence, into a
state of torpid tranquillity. The universal conviction that the dictator’s
power was irresistible had quelled all further heavings of the spirit of
discontent. Dolabella had been gratified with a command in the late
campaign; while others, in whose fidelity and military skill he could
rely, had been left behind to overawe disaffection. The most
illustrious of the nobility having now no occasion to remain at Rome
for the sake of paying court to a jealous ruler, had retired generally
to their country seats; but Cicero seems to have feared giving
occasion for distrust if he withdrew himself from the broad eye of
public observation. He occupied himself, however, in his philosophical
studies, and could rejoice that he had never, like so many of his
contemporaries when plunging into the excitements of political life,
abandoned the literary pursuits common to them in youth. While he
still regarded the contest in Africa with the sentiments of a true
republican, he confessed with a sigh that though the one cause was
assuredly the more just, yet the victory of either would be equally
disastrous. He probably held aloof from the proceedings of the
servile senate, which occupied itself during the months of Cæsar’s
absence in devising new honours for his acceptance. First of all it
decreed the religious ceremony of a thanksgiving of forty days,
being twice the term to which the compliance of popular gratitude
had ever previously extended, and it was by the length of the
observance that the honour was estimated. Next it appointed that
the victor’s triumphal car should be drawn by horses of white, the
sacred colour, and that the number of his attendant lictors should be
doubled. He was to be requested to undertake the office of censor
for three years, under a new title, which should not remind the
citizens too closely of the times of republican liberty, that of
præfectus morum, or regulator of manners. The changes which the
revolutionary storm had effected in the condition of so many of the
citizens justified a resort to the old constitutional resource for
purging the senate of scandalous or impoverished members, and
infusing new blood into its veins.
The most substantial of all these tributes to Cæsar’s ascendency
was the decree by which he was appointed dictator for a period of
ten years; for thus the initiative of legal measures was united in his
hands with the command of the legions both at home and abroad.
Other specious honours, in the taste of the times, were accumulated
upon him. His chair was to be placed between those of the consuls
in the assembly of the senate; he was to preside and give the signal
in the games of the circus; and his figure in ivory was to be borne in
procession among the images of the gods, and laid up in the Capitol,
opposite the seat of Jupiter himself. A statue was to be erected to
him in bronze, standing upon a globe, with the inscription, “Cæsar
the demi-god.” His name was to be engraved on the entablature of
the Capitol, in the place of that of Catulus, its true restorer. The
historian who recounts these honours assures us that many others
besides these were offered; he has only omitted to specify them
because Cæsar did not think fit to accept them. It is difficult to
imagine to what lower depth of obsequiousness the senate could
have descended, or what higher dignities the conqueror would have
rejected.

CÆSAR’S TRIUMPHS
 The time had now arrived for the celebration of the Gallic triumph,
which had been so long postponed. In the interval, the imperator’s
victories had been multiplied, and the ranks of his veterans had been
recruited by fresh enlistments; so that every soldier who had shared
in his later perils and successes demanded the reward of
participating in his honours. Cæsar claimed not one, but four
triumphs: the first, for his conquest of the Gauls; the second for his
defeat of Ptolemy; another, for his victory over Pharnaces; and the
last, for the overthrow of Juba. But he carefully avoided all reference
to what were in reality the most brilliant of his achievements. In
Spain and Thessaly he had routed the disciplined legions of his own
countrymen; but their defeat brought no accession of honour or
territory to the republic. The glory it reflected on the victor was
dubious and barren. The four triumphs were celebrated, with
intervals of a few days between each, that the interests of the public
might not pall with satiety. The first procession formed in the
Campus Martius, outside the walls of the city. It defiled through the
triumphal gate at the foot of the Capitoline Hill, and crossed the
deep hollow of the Velabrum and Forum Boarium, on its way to the
Circus Maximus, which occupied the valley between the Palatine and
Aventine. In passing through the Velabrum, the chariot in which the
imperator stood, happened to break down; a mischance which so
affected him that he never afterwards, it is said, ascended a vehicle
without repeating a charm.
The long procession wound round the base of the Palatine,
skirting the Aventine and Cælian hills, to the point where the arch of
Constantine now stands. There it began the ascent of the gentle
slope which separates the basin of the Colosseum from that of the
Roman Forum. It followed the same track which now leads under
the arch of Titus, paved at this day with solid masses of hewn stone,
which may possibly have re-echoed to the tramp of Cæsar’s legions.
Inclining a little to the right at the point where it gained the summit
of the ridge and looked down upon the comitium and rostra, in the
direction of the Capitol, it passed before the spot where the temple
of Julius was afterwards built; thence it skirted the right side of the
Forum, under the arch of Fabius, till it reached a point just beyond
the existing arch of Severus, where the two roads branched off, the
one to the Capitoline temple, the other to the Mamertine prison.
Here it was that Cæsar took the route of triumph to the left, while
Vercingetorix was led away to the right, and strangled in the
subterranean dungeon. The Gallic hero doubtless met with firmness
and dignity the fate to which he had so long been doomed, while his
conqueror was exhibiting a melancholy spectacle of human infirmity,
crawling up the steps of the Capitol on his knees, to avert, by an act
of childish humiliation, the wrath of the avenging Nemesis. The next
instance of similar degradation recorded is that of the emperor
Claudius, who being corpulent and clumsy performed the ungraceful
feat with the support of an arm on either side. The practice was
probably of no unusual occurrence, and was deeply rooted, we may
believe, in ancient and popular prejudices. A remnant of it still exists,
and may be witnessed by the curious, even at the present day, on
the steps of the Ara Cœli and at the Santa Scala of the Lateran.
The days of triumph which succeeded passed over with
uninterrupted good fortune. The populace were gratified with the
sight of the Egyptian princess Arsinoe led as a captive at the
conqueror’s wheels; but she was spared the fate of the Gallic
chieftain out of favour to her sister, or perhaps out of pity to her sex.
The son of the king of Numidia who followed the triumphal car was
also spared, and lived to receive back his father’s crown from
Augustus. Though Cæsar abstained from claiming the title of a
triumph over his countrymen, he did not scruple to parade their
effigies among the shows of the procession. The figures or pictures
of the vanquished chiefs were carried on litters, and represented the
manner of their deaths. Scipio was seen leaping desperately into the
sea; Cato plunging the sword into his own bowels; Juba and Petreius
engaged in mortal duel; Lentulus stabbed by the Egyptian assassin;
Domitius pierced perhaps in the back, in token of his flight. The
figure of Pompey alone was withheld for fear of the commiseration it
might excite among the people whose favourite he had so lately
been. Nor, as it was, were the spectators unmoved. Upon the
 unfeeling display of Roman defeat and
disaster they reflected with becoming
sensibility. But the pictures of Achillas and
Pothinus were received with unmingled
acclamations, and loud was the cry of scorn
at the exhibition of Pharnaces flying in
confusion from the field. After all, the most
impressive part of the ceremony must have
been the appearance of the rude veterans
whose long files closed the procession. With
what ignorant wonder must the children of
Gaul and Iberia, of Epirus and Africa, have
gazed at the splendour of the city, of which
the fame resounded in their native cabins!
What contempt must they have felt for the
unarmed multitudes grinning around them!
How reckless must they have been of the
A Sacrificator dignity of the consuls and senators, they
who claimed the license of shouting derisive
songs in the ears of their own commander!
Little did they think that grave historians would sum up their coarse
camp jokes in evidence against the fame of their illustrious leader;
still less did they dream of the new power which the military class
was thenceforth to constitute in the state. Rome in fact was their
own; but it was a secret they were not yet to discover.
The satisfaction of his armed supporters, however, was the first
condition on which the supreme power of the dictator must
henceforth be maintained in the city. It was a matter, indeed, of
hardly less importance to secure the good humour of the urban
population. While the soldiers receive each a donative of twenty
thousand sesterces, the claims of the much larger multitude of the
free citizens were not undervalued severally at four hundred;
especially as they received the additional gratification of one year’s
remission of house rent. It does not appear how this indulgence
differed from that for which Cælius and Dolabella had raised their
commotions; but the dictator had so strenuously resisted every
attempt to set aside the just claims of creditors on all previous
occasions, that it can hardly be doubted that in this case he gave the
landlords compensation from the public treasury. The mass of the
citizens was feasted at a magnificent banquet, at which the Chian
and Falernian wines, the choicest produce of Greece and Italy,
flowed freely from the hogshead, and towards which six thousand
lampreys, the most exquisite delicacy of the Roman epicure, were
furnished by a single breeder. The mighty multitude reclined before
twenty-two thousand tables; each table having its three couches,
and each couch, we may suppose, its three guests; so that the
whole number feasted may have amounted to nearly two hundred
thousand. When Cæsar undertook the functions of his censorship,
the number of recipients of the public distributions of corn was
estimated at 320,000. Upon a scrutiny into their claims as genuine
and resident citizens, he was enabled to strike off as many as
150,000 from this list. Adding to the remainder the senators and
knights, and the few wealthy individuals who might have scorned to
partake of a state provision, the sum will correspond pretty
accurately with the number of the imperial guests as above
computed.
The public shows with which these gratifications were
accompanied were carried out on a scale of greater magnificence
than even those recently exhibited by Pompey. There was nothing in
which the magistrates of the republic vied more ostentatiously with
one another than in the number of wild beasts and gladiators which
they brought into the arena. The natural taste of the Italian people
for shows and mummery degenerated more and more into an
appetite for blood; but in this, as in every other respect, it was
Cæsar’s ambition to outdo his predecessors, and the extraordinary
ferocity and carnage of the exhibitions which he complacently
witnessed excited a shudder even in the brutal multitude. The
combatants in the games of the Circus were either professional
gladiators, who sold their services for a certain term of years, or
captives taken in war, or lastly public criminals. But Cæsar was,
perhaps, the first to encourage private citizens to make an exhibition
of their skill and valour in these mortal combats. He allowed several
men of equestrian rank, and one the son of a prætor, to demean
themselves in the eyes of their countrymen by this exposure to the
public gaze. It was only when a senator named Fulvius Setinus
asked permission thus to prostitute his dignity, that the dictator was
at last roused to restrain the growing degradation.
If the people of Rome were shocked at the bloodshed which they
were invited to applaud, it seems that they were offended also at
the vast sums which were lavished on these ostentatious spectacles.
They would have preferred, perhaps, that the donative to
themselves should have been greater, and the soldiers even
exhibited symptoms of discontent and mutiny in consequence. No
instance of Cæsar’s profuse expenditure excited greater admiration
than his stretching a silken awning over the heads of the spectators
in the Circus. This beautiful material was brought only from the
farthest extremity of India, and was extremely rare and precious at
Rome at that time. Three centuries later it was still so costly that a
Roman emperor forbade his wife the luxury of a dress of the finest
silk unmixed with a baser fabric. But a more permanent and worthy
object of imperial expenditure was the gorgeous Forum of which
Cæsar had long since laid the foundation with the spoils of his Gallic
Wars. Between the old Roman Forum and the foot of the Quirinal, he
caused a large space to be enclosed with rows of marble corridors,
connecting in one suite halls of justice, chambers of commerce, and
arcades for public recreation. In the centre was erected a temple to
Venus the ancestress, the patroness for whom Cæsar had woven a
breastplate of British pearls, and whose name he had used as his
watchword on the days of his greatest victories. He now completed
the series of his triumphal shows by the dedication of this favourite
work. It remained for centuries a conspicuous monument of the
fame and magnificence of the first of the Cæsars. His successors
were proud to cluster new arches and columns by its side, and
bestowed their names upon the edifices they erected in connection
with it. Finally, Trajan cut through the elevated ridge which united
the Capitoline with the Quirinal, and impeded the further extension
of the imperial forums. He filled the hollow with a new range of
buildings, occupying as much ground as the united works of his
predecessors in this quarter. The depth of his excavation is indicated,
it is said, by the height of the pillar which bears his name.

THE LAST CAMPAIGN

Our review of the dictator’s proceedings in

[46-45 b.c.] the discharge of his civil functions must be
postponed, but only for a moment, to relate the
short episode of his last military exploit. The despatches of his
lieutenants in Spain represented that province as rapidly falling into
the hands of the republican faction. Varus and Labienus had escaped
from Africa, and joined the standard under which Scapula
marshalled the disaffected legions in Spain. Cneius Pompeius had
also issued from his retreat in the Balearic Isles, and as soon as he
appeared in their camp every chief of the oligarchy waived his own
pretensions to the command in deference to the man who
represented the fame and fortunes of their late leader. Yet Scapula
had the confidence of the soldiers, Labienus was an officer of tried
ability and reputation, and Varus had at least held the highest
military commands, while Cneius himself was personally unknown to
the legions in Spain, and his only achievement in war had been a
dashing naval exploit. So cowed by its repeated reverses was the
spirit of the old Roman party, which had revived for a moment in
Africa with vain exultation at finding itself relieved from the
ascendency of its own military champion. Cneius, on his part, seems
to have regarded the renewed contest in the light of a private
quarrel. His war-cry was not “Rome,” “Liberty,” or “The Senate,” but
“Pietas,” “Filial Duty.”
The disaffection among Cæsar’s soldiers had become widely
spread; a large body of them had enrolled themselves under their
new leaders; their numbers had been augmented by provincial
enlistments; even slaves had been drafted into the ranks; while the
cities and states of the peninsula lent their aid more or less openly
to the cause. It was not in the remoter parts of the province or
among the half-subdued native principalities, but in the centre of
Roman influence and civilisation, in Corduba itself, that the standard
of the adventurers was unfurled. Cæsar had completed the
ceremonies of his quadruple triumph, and was deeply engaged in
the arduous task of legislation for the new system of government
which he had undertaken to raise, when he found it necessary to
postpone every other occupation to meet his enemies once more in
arms. So uncertain and tedious was the navigation of those days
that he may have chosen the land route across the Alps and
Pyrenees, for the sake of reaching his destination with greater
speed.[123]
The details of the campaign into which he immediately plunged
are given, but very obscurely, in the last of the series of
contemporary memoirs which have hitherto been our guides
throughout the military history of the period. In point of composition
it betrays less literary accomplishment than any of its kindred works.
The rude soldier who seems to have been its author had no
hesitation in recording in their undisguised enormity the cruelties
which disgraced the conduct of both parties. Cæsar’s character for
humanity suffers more in this than in any other contemporary
narrative of his actions. The campaign was, indeed, a series of
butcheries on either side, but Cneius was, perhaps, the most
savagely ferocious of all the captains of the civil wars. The scene of
the last act of Roman liberty was laid in the valley of the
Guadalquivir and the defiles of the Sierra de Tolar. After a variety of
desultory movements, of which we obtain from the narrative only an
indistinct notion, we find the rival armies at last drawn up in hostile
array on the field of Munda. Cæsar was this time superior in
numbers, and especially in cavalry; but the enemy was well posted,
and fought well: never, it is said, was the great conqueror brought
so near to defeat and destruction.b
 “When the armies were going to close, Cæsar, seeing his men go
on but coldly and seem to be afraid, invoked all the gods,
beseeching them with hands lifted up to heaven, not to let the lustre
of so many glorious actions be darkened in one day, and running
through the ranks, encouraged his soldiers, taking off his head-piece
that he might be better known. But do what he could, he could not
raise their spirits, till snatching a buckler out of a soldier’s hand, he
said to the tribunes who were about him, ‘This shall be the last day
of my life, and of your engagement in the war.’ And at the same time
made furiously towards the enemy; he had scarce advanced ten feet
but he had above two hundred darts thrown at him, some of which
he avoided by bending his body, and others received on his buckler,
when the tribunes ran with emulation to get about him, and the
whole army thereupon charging with all their fury, they fought all
day with divers advantage, and at length towards the evening the
victory fell to Cæsar, and it is reported that hereupon he was heard
to say these words, ‘that he had often fought for victory, but that
now he had fought for life.’
“After the defeat, Pompeius’ men flying into Corduba, Cæsar, to
prevent their escape thither, lest they should rally and renew the
fight, caused the place to be invested by the army, where the
soldiers being so tired that they could not work in the
circumvallation, heaped up together the bodies and armour of the
slain, which they kept piled up with their javelins stuck into the
ground, and lay all night under that kind of rampire. Next morning
the city was taken. Of Pompeius’ captains, Scapula setting up on a
pile of wood burned himself; the heads of Varus, Labienus, and
other persons of quality were brought to Cæsar. As for Pompeius, he
fled from the battle with a hundred and fifty horse, bending his
course towards Carteia where his fleet lay; he entered the port in a
litter, and in the habit of a private man. But seeing the seamen had
likewise lost all hopes, he threw himself in a little boat, in which as
he was going out to sea, his foot tangling in the cordage, one of his
people going to cut the rope, by mischance cut his heel, so that to
cure his wound he was forced to go ashore at a small village, where
hearing that Cæsar’s horsemen were coming, he took his flight
through a country covered with thorns and briers, which added to
his wound, so that being tired and sitting down at last under a tree,
he was found by those who gave him chase, and slain, generously
defending himself; his head was carried to Cæsar, who caused it to
be buried. Thus [says Appian] was this war ended by one only fight
and contrary to the opinion of all the world.”g
Of all the leaders of the senatorial party, Sextus Pompeius was
now the only survivor. He had made his escape from the field of
Munda, and had an asylum in the wildest districts of the Hither
Province. He had nothing to hope from the clemency of the
conqueror, who had shown unusual bitterness against his family by
the confiscation of their patrimonial estates, and was now preparing
to celebrate his triumph over them as foreigners and enemies of the
state. Thus driven to despair, he infused new spirit into the
predatory habits among the tribes among whom he had taken
refuge, and continued to defy the power of the provincial authorities.
Cæsar occupied himself for some months in reconstituting the
government of Spain, taking precautions for the entire subjugation
of the party which had shown such vitality in that quarter. The battle
of Munda was fought on the seventeenth of March, but the dictator
was not at liberty to return to Italy till September, after an absence
of ten months.
The hostile attitude of the last of the Pompeians in Spain was not
the only exception to the tranquillity which prevailed generally
throughout the empire. In Gaul the Bellovaci had risen in arms; but
this movement was expeditiously repressed by Decimus Brutus, the
proconsul of the newly conquered province. In the extreme East,
however, the republican party still continued to make head, under
the leadership of Cæcilius Bassus. Their champion was an obscure
knight, and their forces were insignificant, consisting principally of
two legions which Bassus had seduced from their allegiance to
Sextus Cæsar, the commander to whose care Syria had been
entrusted by his kinsman. But the proximity of the Parthians, ever on
the watch for an opportunity to wound the sides of their great rivals,
rendered any movement in this quarter formidable. Sextus Cæsar
was murdered by his soldiers, and Bassus took possession of the city
of Apamea, which, with the assistance of the national enemies, he
continued to keep against the petty attempts which were made to
dislodge him. The dictator kept his eye upon him, and already
meditated his destruction; but for the present he was content to
leave his temerity unpunished, while he applied himself to the
consolidation of his power by bold and comprehensive legislation at
home.b

THE LAST TRIUMPH

On the 13th of September, 45, the dictator

[45 b.c.] appeared once more at the gates of Rome, but
he did not triumph till the commencement of
October. His victory was represented as gained over the Iberians;
the miserable outcasts whom Cneius had banded together were all
confounded together under the common title of strangers and
enemies. Two of the dictator’s lieutenants, Fabius, and Pedius who
was also his kinsman, were allowed the honour of separate
triumphs. These ceremonies were followed as usual with games and
festivals, which kept the populace in a fever of delight and
admiration. They had complained that among the numerous
spectacles offered to their view each citizen could witness only a
portion, while to the foreigners who flocked to this great feast of
nations, the dramatic entertainments had been unintelligible. The
games were now multiplied in various quarters of the city, while
plays were represented in different languages for the benefit of
every people. The subjects of the empire had entered Rome as
conquerors in Cæsar’s train, and thus he inaugurated the union of
the capital with the provinces. Kings and commonwealths sent their
ambassadors to this mighty congress of nations. Among them were
the Moors and the Numidians, the Gauls and the Iberians, the
Britons and the Armenians, the Germans and the Syrians. The Jews,
insulted by Pompey and rifled by Crassus, offered their willing
homage to the champion who alone of all the Romans had spoken to
them in the language of kindliness and respect. Cleopatra the queen
of Egypt came, her crown in her hand, offering her treasures and
her favours to her admirer and preserver. All in turn had trembled at
the official caprices of the Roman knights, and Cæsar could afford
them perhaps no sweeter revenge, nor represent to them more
vividly the extent of his power, than in degrading before their faces
these petty tyrants of the provinces. He compelled one of them,
named Laberius, who was also a dramatic composer, to enact one of
his own comic pieces, that is, to dance and sing upon the stage
before the concourse of citizens and strangers. “Alas!” said the
wretched man in his prologue, “after sixty years of honour I have
left my house a knight, to return to it a mime. I have lived one day
too long.” Cæsar restored to him the golden ring of knighthood,
forfeited by this base but compulsory compliance. He presented him
also with a large sum of money, to show perhaps more completely
the prostration of his order.
Such trifling persecutions, whether personal or political in their
objects, are undoubtedly pitiable enough. But it is Cæsar’s glory that
his arm fell heavily upon none of his fellow-citizens. The nephew of
Marius forgot the banishment of his uncle, the ruins of Carthage,
and the marshes of Minturnæ; the avenger of the Sullan revolution
scorned to retaliate the proscriptions; the advocate of Cethegus and
Lentulus refrained from demanding blood for blood. It is worth
remarking that Cicero, the most humane perhaps of his own party,
the most moderate in sentiments, the fairest estimater of men and
measures, could hardly persuade himself of the possibility of Cæsar
abstaining from massacre. Such was the wise man’s reading of the
history of his countrymen; and when at last he found that the
conqueror meditated no such use of his victory, his heart, we fear,
still remained untouched, and he never, perhaps, renounced the
secret hope that Cæsar’s opponents would prove less merciful than
himself.
 Nor was the conqueror’s clemency confined to sparing the lives of
his opponents. He refrained from confiscation which had been wont
to accompany the edicts of his predecessors. The wealth indeed
which was poured into Rome from the tribute of so many new
subjects, and the plunder of so many temples, rendered it more easy
to practise this unusual liberality. It was ungenerous perhaps to
make the estates of his great rival the chief exception to this rule of
moderation. But Cæsar intended to brand as rebels to constituted
authority the men who renewed the strife after Thapsus, and this
confiscation was meant, not as an insult to the dead, but as a
punishment of the living opponent. The name of the Great Pompey
had already passed into the shrine of history, and the victor was
proud of closing the fasti of the republic with so illustrious a title. Far
from approving the precipitation of his flatterers in removing the
statues of Pompey and Sulla, he caused them to be restored to their
places in front of the rostra, among the effigies of the noblest
champions of the free state. Towards the institutions of the
commonwealth he evinced a similar spirit of deference. He sought
no new forms under which to develop his new policy. Sulla had
attempted to revive the aristocratic spirit of the ancient constitution
by overthrowing the existing framework of the laws; but the popular
dictator, in laying the foundation of a more extensive revolution,
studied to preserve it intact. While making himself an autocrat in
every essential exercise of power, he maintained, at least in outward
seeming, all the institutions most opposed to autocracy, the senate,
the comitia, and the magistracies. But he had long before said that
the republic was no more than a shadow, and these very institutions
had long been merely the instruments by which tyrants had worked
out the ends of their selfish ambition.
Cæsar now was fully aware that he could sway the Roman world
unchecked by the interference of a senate, two-thirds of which
perhaps were nominees of his own. Under the sanction of an organic
law he had raised the number of the assembly to nine hundred, thus
degrading the honour by making it cheap; and he still more
degraded it in the eyes of the proudest of the citizens by pouring