Cluster Comput (2016) 19:1787–1800

DOI 10.1007/s10586-016-0623-4

SOCCER: Self-Optimization of Energy-efficient Cloud Resources

Sukhpal Singh1 · Inderveer Chana1 · Maninder Singh1 · Rajkumar Buyya2

Received: 2 July 2016 / Revised: 29 July 2016 / Accepted: 18 August 2016 / Published online: 1 September 2016
© Springer Science+Business Media New York 2016

Abstract Cloud data centers often schedule heterogeneous Keywords Cloud computing · Energy · Resource schedul-
workloads without considering energy consumption and ing · Autonomic computing · Self-optimization · Green
carbon emission aspects. Tremendous amount of energy con- computing
sumption leads to high operational costs and reduces return
on investment and contributes towards carbon footprints to
the environment. Therefore, there is need of energy-aware 1 Introduction
cloud based system which schedules computing resources
automatically by considering energy consumption as an Cloud offers three types of services (Infrastructure, Plat-
important parameter. In this paper, energy efficient auto- form and Software) using pay per use model. Data centers
nomic cloud system [Self-Optimization of Cloud Computing are the backbone of the modern economy; from the server
Energy-efficient Resources (SOCCER)] is proposed for rooms that power small-sized to medium-sized organizations
energy efficient scheduling of cloud resources in data centers. to the enterprise data centers that support corporations and
The proposed work considers energy as a Quality of Service the server farms that run cloud computing services hosted
(QoS) parameter and automatically optimizes the efficiency by Amazon, Facebook, Google, and others [1]. However,
of cloud resources by reducing energy consumption. The the explosion of digital content, big data, e-commerce, and
performance of the proposed system has been evaluated in Internet traffic is also making data centers one of the fastest-
real cloud environment and the experimental results show growing consumers of electricity in developed countries.
that the proposed system performs better in terms of energy Further, cloud data centers, provide effective and reliable
consumption of cloud resources and utilizes these resources infrastructure services to the end users [2]. Presently, cus-
optimally. tomer satisfaction and performance is increased by the data
centers without considering energy consumption in those dat-
acenters which leads to high operational cost and reduces
B Sukhpal Singh return on investment. Many governments have also imposed
[email protected] constraints to reduce the carbon footprints which effects envi-
Inderveer Chana ronment. IT companies (Microsoft, Google, Amazon, IBM
[email protected] etc.) are increasing their data centers every year to provide
Maninder Singh services to the cloud users in a better way [3,4]. Due to large
[email protected] energy consumption, temperature increases gradually which
Rajkumar Buyya leads to the failure of the system and violates the service
[email protected] level agreement (SLA). Literature reports that data center
1 Computer Science and Engineering Department,
infrastructure generates over 70 % of total heat generated [5].
Thapar University, Patiala, Punjab 147004, India Another reason of wastage of energy is resources are run-
2 CLOUDS Lab, Department of Computing and Information
ning in idle or underutilized state. Energy efficient resource
Systems, The University of Melbourne, Melbourne, VIC, scheduling in cloud is a challenging job and the scheduling
Australia of appropriate resources to cloud workloads depends on the

123
1788 Cluster Comput (2016) 19:1787–1800

Quality of Service (QoS) requirements of cloud applications in Sect. 3. Section 4 describes the experimental setup and
and energy consumption of computing resources [6]. Energy presents the results of evaluation. Section 5 presents conclu-
saving in case of heterogeneous cloud workloads is very dif- sions and future scope.
ficult to improve. Therefore, there is need of cloud based
system which schedules computing resources automatically
by considering energy consumption as a significant aspect. 2 State-of-the-art
Scheduling of resources in cloud is an important part
of resource management system. Mapping of cloud work- Scheduling of resources in cloud has been done through
loads to appropriate resources is mandatory to improve QoS different techniques as reported by literature but efficiency
parameters like time, cost, energy consumption etc. In our saving is an important factor that is difficult to optimize
earlier work [1–4], we have identified various research issues automatically. To reduce energy consumption, researchers
related to QoS and SLA for cloud resource scheduling and proposed the concept of VM (virtual machine) consolida-
have developed a QoS based resource provisioning technique tion to detect overload, under-load and VM selection [5–7].
(Q-aware) to map the resources to the workloads based on Wang et al. [8] described the overview of data center net-
user requirements described in the form of SLA. Further, works for cloud computing. Further, detailed descriptions of
QoS based Resource Scheduling Framework (QRSF) has virtualized infrastructure, physical architecture and dynamic
been proposed, in which provisioned resources have been circuit network (DCN) routing have been discussed.
scheduled by using different resource scheduling policies Salehi et al. [9] proposed Preemption-Aware Energy Effi-
(cost, time, cost-time and bargaining based). Based on QoS cient (PAEE) technique for virtualized datacenters which
requirements, scheduler finds and maps the resources and adjusts the energy consumption based on user performance
workloads. Resource scheduling in previous work [2–4] has requirements and reduces SLA violations. PAEE reduces
been done in following steps: (i) understand the expectations energy consumption up to 18 % but it considers only homoge-
and requirements of cloud user, (ii) analyze and cluster the nous workloads. Ren et al. [10] proposed provably-efficient
workloads through k-means clustering algorithm, (iii) find online scheduling algorithm for geographically distributed
the required number of resources, (iv) map the resources and datacenters which optimizes the energy cost and fairness
workloads and (v) schedule and execute the workload on among different organizations subject to queueing delay con-
appropriate resources with minimum time and cost. QRSF straints and it reduces energy consumption which is closed
framework executes the workloads without self-optimization to optimal offline algorithm. Pelley et al. [11] developed
of resources. To incorporate self-optimization, QRSF has mechanisms to better utilize installed power infrastructure,
been further extended by proposing Energy-aware Auto- reducing reserve capacity margins and avoiding performance
nomic Resource scheduling TecHnique (EARTH) [1], in throttling. Further, it reduces reserve capacity requirements
which IBM’s autonomic computing concept has been used to to tolerate a single power distribution unit (PDU) failure. In
schedule the resources automatically by optimizing energy addition, power routing is proposed to schedule workload
consumption where user can easily interact with the system dynamically across different servers. Urgaonkar et al. [12]
using available user interface. But EARTH can execute only proposed an Lyapunov optimization technique based online
homogenous cloud workloads and the complexity of resource control algorithm that can optimally exploit these devices
scheduling in EARTH increases with the increase of num- to minimize the time average cost. It operates without any
ber of workloads. To address this issue, proposed system knowledge of the statistics of the workload or electricity cost
(SOCCER) clusters the heterogeneous cloud workloads and processes, making it attractive in the presence of workload
executes them with minimum energy consumption. and pricing uncertainties. Shen et al. [13] formulated sev-
The motivation of this paper is to design energy efficient eral stochastic optimization models for trading off between
autonomic cloud computing system called Self Optimization energy footprints and QoS associated with server consolida-
of Cloud Computing Energy-efficient Resources (SOCCER) tion in cloud computing data centers. Further, they consider
for effective scheduling of resources which considers energy finite service times with uncertain workloads at each period
consumption as a QoS parameter. The main aim of this and minimize the expected energy consumption.
research work is: (i) to propose an autonomic resource Changtian et al. [14] described DVS (dynamic voltage
management technique for execution of heterogeneous work- scaling) based energy aware technique to execute workloads
loads by considering generic property of self-management, with minimum execution time and energy consumption. Fit-
(ii) to optimize the energy consumption and (iii) to imple- ness function is defined based on methods of double and
ment and perform evaluation in a real cloud environment for unify fitness and genetic algorithm is used to identify the
clustered heterogeneous workloads. The rest of the paper is resources with minimum energy consumption. Ma et al.
organized as follows. Section 2 presents related work of exist- [15] described control dependence graph based energy aware
ing energy-efficient systems. Proposed system is presented resource scheduling technique to execute the HPC applica-

123
Cluster Comput (2016) 19:1787–1800 1789

Table 1 Comparison of
Framework Mechanism Workload type Clustering of
SOCCER with existing
workloads
frameworks
PAEE [9] Non-autonomic Homogenous ×
ECS [16] Non-autonomic Homogenous ×
GreenDCN [21] Non-autonomic Homogenous ×
GreFar [22] Non-autonomic Homogenous ×
CMSS [23] Non-autonomic Homogenous ×
√
SOCCER Autonomic Homogenous and
heterogeneous

tions in distributed environment within deadline with least scheduling algorithm (GreFar) is proposed to optimize the
energy consumption. Design approximation and traditional energy cost and fairness among different organizations sub-
multiprocessor scheduling algorithms are extended to formu- ject to queueing delay constraints. Liu et al. [23] proposed
late the problem after analysis and completion of worst-case industrial cluster oriented Cloud Manufacturing Service Sys-
performance. Further based on energy consumption and tem (CMSS) in order to fulfill the real time designing and
desired deadline of tasks, pricing scheme is designed for manufacturing information interaction among the collabora-
their execution. Kim et al. [16] proposed energy credit sched- tive partners in an industrial cluster area. Further, lightweight
uler (ECS) used to estimate the consumption of power in application of CMSS is designed to analyses the services
VM based on the number of workloads executed on VM. using virtual cloud environment. Mashayekhy et al. [24]
Scheduling algorithm for virtual environment is designed designed an auction-based online mechanism for VM pro-
based on this estimation model to execute the tasks on com- visioning, allocation, and pricing in clouds that considers
puting resources based on minimum energy consumption several types of resources. Further, it allocates VM instances
and budget and implemented in Xen virtualization system to selected users for the period they are requested for,
and it reduces energy consumption with minimum error rate. and ensures that the users will continue using their VM
Chen et al. [17] proposed holistic workload based resource instances for the entire requested period. Xu et al. [25]
scheduling policy for geographical distributed data cen- studied traffic-aware resource provisioning mechanisms for
ters to improve energy efficiency and MinBrown (workload distributed clouds and examines cloud traffic characteristics
scheduling technique) is designed and consider constraints and optimizations produced fine-grained traffic-awareness
like availability of green energy and cooling power. approaches that can more efficiently reduce energy costs
Wang et al. [18] proposed virtualization based unified opti- for distributed clouds with dynamic, diverse traffic. SOC-
mization framework to improve the energy efficiency of data CER (self optimization of cloud computing energy-efficient
center networks by using the concept of multipath routing resources) has been compared with existing frameworks as
protocol and hierarchical feature of the topology. Wang et described in Table 1.
al. [19] proposed an efficient energy saving technique for None of the existing research work considers heteroge-
data center networks by scheduling and routing “deadline- neous cloud workloads, clustering of workloads and auto-
constrained flows” where the transmission of every flow has nomic management of resources. In addition; SOCCER
to be accomplished before a rigorous deadline, being the needs to consider the basic features of cloud computing in
most critical requirement in production data center networks. order to execute the heterogeneous cloud workloads auto-
Shu et al. [20] proposed architecture of cloud-integrated matically with minimum energy consumption and maximum
cyber-physical systems to execute complex industrial appli- energy efficiency.
cation in a controlled manner. Wang et al. [21] proposed an
energy efficient framework (GreenDCN) by assigning vir- 2.1 Our previous contributions
tual machines to servers to reduce the amount of traffic and
to generate favorable conditions for traffic engineering. Fur- To incorporate self-optimization, QRSF [4] has been further
ther, GreenDCN reduces the number of active switches and extended by proposing Energy-aware Autonomic Resource
balance traffic flows which improves energy efficient of data scheduling TecHnique (EARTH) [1], in which IBM’s auto-
center network. nomic computing concept has been used to schedule the
Polverini et al. [22] explored the benefit of electricity resources automatically by optimizing energy consumption
price variations across time and locations and discussed the where user can easily interact with the system using avail-
problem of scheduling batch jobs to multiple geographically- able user interface. Architecture of EARTH is shown in
distributed data centers. Further, provably-efficient online Fig. 1.

123
1790 Cluster Comput (2016) 19:1787–1800

To monitor energy Manage alerts Select an Action Execute Action

Monitor Analyze Plan Executor

Fuzzification Inference Defuzzification

Interface Engine Interface
Crisp I/P Crisp O/P
Fuzzy Input Fuzzy Output

Sensors
Effectors
To Measure Energy
To Transfer new rules to other nodes
Fuzzy Rule Base

EARTH Knowledge Base

Fig. 1 Energy-aware autonomic resource scheduling framework [1]

2.2 EARTH execution in this research work for Defuzzification. Note: Detailed exe-
cution of EARTH is described in our previous research work
EARTH used fuzzy logic system to process the data in [1].
an effective way [1]. Fuzzy inputs include Workload Wait- But EARTH can execute only homogenous cloud work-
ing Time (WWT), Workload Execution Time (WET) and loads and the complexity of resource scheduling in EARTH
Resource Energy Consumption (REC) and fuzzy output is increases with the increase of number of workloads. To
Workload Processing Priority (WPP). All the four variables address this issue, proposed system (SOCCER) clusters the
are changing continuously due to this, and hence these vari- heterogeneous cloud workloads and executes them with min-
ables are considered. Based on this, fuzzy rule set is created to imum energy consumption.
define the behavior of fuzzy system and setting the relation-
ship among inputs and outputs. Three membership functions 3 SOCCER: Self-Optimization of Cloud
are considered for every input and output variables: Low, Computing Energy-efficient Resources
Medium and High. Based on these input and output variable,
inference engine is making decisions. Value of member- SOCCER focuses on energy efficient scheduling of comput-
ship functions can be changed based on the requirements ing resources in virtual data centers for execution of both
and conditions of every workload. After the inputs and out- homogenous and heterogeneous cloud workloads. SOCCER
puts of a fuzzy system are selected, they must be partitioned focuses on autonomic execution of clustered heterogeneous
into appropriate conceptual categories. Based on selected cloud workloads in order to improve energy efficiency.
inputs and outputs of the fuzzy system, member functions Finally, we have validated SOCCER using real cloud envi-
are created for better representation of relationship among ronment and measured the value of energy consumption of
input and output variables. Each of these categories actually different clusters of workloads. This research work mainly
represents a fuzzy set on a given input or output domain. focuses on one important aspect of self-optimization i.e.
WWT, WET and REC are antecedents and WPP is conse- energy consumption.
quent. Three membership functions consider for every three
inputs. Fuzzification is used to find the degree of truth for 3.1 Objective function
every rule, membership function defined on every input vari-
able is applied to their actual value. We used most popular The goal of cloud provider is to minimize the actual energy
operator “AND” operator for fuzzy implementation. This consumption. The cloud workload will be executed only
function returns the lowest value of among these values when the actual energy consumption denoted as EnCactual
entered. Defuzzification is used to convert the value of fuzzy is less than the threshold value of energy consumption (E t ).
output into crisp output value. We used MAXIMUM method For a particular cloud workload, the information on its energy

123
Cluster Comput (2016) 19:1787–1800 1791

consumption and processor utilization can be used to mea- presented as ru(t). For a resource rt at given time t, the
sure the energy consumption of resources for execution of resource utilization ResUt is defined as (Eq. 4):
heterogeneous cloud workloads.
energy consumption is calculated using (Eqs. 1– 5). The

n
energy model is devised on the basis that resource uti- ResUt = EnCt,i (r u(t))dt (4)
lization has a linear relationship with energy consumption i=1
[1,2]. Energy Consumption (EnC) of using resources can
be expressed as the following formula (Eq. 1):
where n is the number of cloud workloads running at time t.
The actual energy consumption EnCactual of a resource r u t
EnC = EnC Datacenter + EnC T ransceiver s + EnC Memor y
at given time t is defined as (Eq. 5):
+EnC E xtra (1)

EnC Datacenter represents the datacenter’s energy consump- EnCactual = (EnCmax − EnCmin ) × ResUt + EnCmin
tion, EnC T ransceiver s represents the energy consumption (5)
of all the switching equipment. EnC Memor y represents the
energy consumption of the storage device. EnC E xtra rep-
where EnCmax is the energy consumption at the peak load
resents the energy consumption of other parts, including
(or 100 % utilization) and EnCmin is the minimum energy
the fans, the current conversion loss and others. The above
consumption in the active/idle mode (or as low as 1 % uti-
formula can be further disintegrated; a cloud computing envi-
lization).
ronment with d datacenters, t  transceivers equipment and a
centralized memory device, its energy consumption can be
expressed as (Eq. 2): 3.2 SOCCER architecture

EnC = d(EnC Pr ocessor + EnC Primar y Storage Architecture of SOCCER is shown in Fig. 2. SOCCER is
+ EnCsecondar ystorage + EnC Mother boar ds based on IBM’s autonomic model [1] that considers four
+EnC N etwor kCar ds ) + t  (EnC H ar dwar e + EnC L AN car ds steps of autonomic system: (1) Monitor, (2) Analyze, (3) Plan
and (4) Execute. SOCCER comprises of following units:

F
+ dconnector s, f + EnC f )
f =0
3.2.1 Monitors (M)
+ (EnC N etwor k Analysis Ser ver + EnC Memor y Manager
+ EnC N etwor k Attached Storage Arrays ) + EnC E xtra (2) Initially, Monitors (M) are used to collect the information
from sensors (Sensors get the information about energy con-
The energy consumed by a transceiver and all its ports can sumption of all the systems working under cloud and update
be defined as: where EnC H ar dwar e is related to the energy the information time to time) for monitoring continuously
consumed by the transceiver, EnC L AN car ds is the energy the value of energy consumption as shown in [Algorithm 1:
consumed by any active network LAN card, EnC f corre- Monitoring Unit (MU)] and transfer this information to next
sponds to the energy consumed by a connector (port) running module for further analysis.
at the frequency f . In the equation (Eq. 2), only the last com-
ponent appears to be dependent on the link frequency while
other components, such as EnC H ar dwar e and EnC L AN car ds
remain fixed for all the duration of transceiver operation.
Therefore, EnC H ar dwar e and EnC L AN car ds can be avoided
by turning the transceiver off or putting it into sleep mode.
EnCt,i is the energy consumption at given time t is defined
in (Eq. 3):

EnCt,i (r ) = q × EnCmax + (1 − q) × EnCmax × r u (3)

where EnCmax is maximum energy consumption while

resource is fully utilized, q is fraction of energy consumed
by idle resource and ru is resource utilization. Resource uti-
lization is change over time and it is function of time and

123
1792 Cluster Comput (2016) 19:1787–1800

Fig. 2 SOCCER architecture

Monitor Manage Select an Execute
QoS alerts Action Action

Monitor Analyze Plan Executor

SOCCER
Autonomic Manager

Provisioning Scheduling Execution

Sensors Resource Information Database Effectors

3.2.2 Analyze and plan (AP) Workload submitted by user to resource provisioner
is stored into bulk of workloads for their execution. All
Analyze and Plan (AP) module starts analyzing the informa- the submitted workloads are analyzed based on their QoS
tion received from monitoring module and makes a plan for requirements described in terms of SLA as shown in
adequate actions for corresponding alert as shown in [Algo- Table 2.
rithm 2: Analyzing and Panning Unit (AU)]. In this step, Workload patterns are identified for better classification of
based on QoS requirements of workload(s), resources are workloads then pattern based clustering of workloads is done.
provisioned, scheduled and executed. It comprises of fol- QoS metrics for every QoS requirement of each workload
lowing subunits: are identified [2–4]. Based on importance of the attribute,
weights for every cloud workload are calculated. After that,
workloads are re-clustered based on k-means based cluster-
ing algorithm for better execution. Final set of workloads
is shown in Table 3. Note: Detailed description of clus-
tering of workloads is described in our previous research
work [4].
Estimate the value of energy consumption of cloud
resources based on their previous statistics of execution avail-
able in Resource Information Database (RID). If the value
of energy consumption of workloads executes within range
[Actual Energy Consumption denoted as EnCactual is less
than the threshold value of energy consumption (E t )] then it
will provision resources otherwise generate alert for analyses
the workload again after reallocation of resources by auto-
nomic manager. After finding the workload priority using
EARTH [1], SOCCER calculates the resource requirements
to check whether the provided resources are sufficient for
Resource provisioning Monitor continually checks the sta- execution of workload (s) are provided or not. If the sufficient
tus of resources provisioned, workloads queued and energy resources are provided then start scheduling of resources for
consumption. The objective of resource provisioner is to pro- workload execution otherwise add new resources from pool
vision the resources for execution of heterogeneous cloud of reserved resources. Resources are again allocated for fur-
workloads without violation of SLA [3]. The workloads ther execution after finding the minimum value of energy
submitted should be executed with minimum energy con- consumption (also less than threshold value). The outcome of
sumption. SOCCER provisions and schedules the resources resource provisioning is set of provisioned resources, which
based on energy consumption to the workloads automatically is stored in RID.
as shown in Fig. 3.

123
Cluster Comput (2016) 19:1787–1800 1793

Fig. 3 Automatic execution of

SOCCER Identify QoS Identify Resource
Requirements Requirements

Resource Resource Resource

Provisioning Scheduling Execution [E]

Generate
[No] < [YES]
Reallocate [M]
Resources Alert [AP]

Table 2 Cloud workloads and

Workload QoS requirements
their QoS requirements
Websites Reliable storage, high network bandwidth, high availability
Technological computing Computing capacity, reliable storage
Endeavour software Security, high availability, customer confidence level, correctness
Performance testing Computing capacity, network bandwidth, latency
Online transaction processing Security, high availability, internet accessibility, usability
E-com Variable computing load, customizability
Central financial services Security, high availability, changeability, integrity
Storage and backup services Reliability, persistence
Productivity applications Network bandwidth, latency, data backup, security
Software/project User self-service rate, flexibility, creative group of
development and testing infrastructure services, testing time
Graphics oriented Network bandwidth, latency, data backup, visibility
Critical internet applications High availability, serviceability, usability
Mobile computing services High availability, reliability, portability

Table 3 Clustering of workloads [2]. (Algorithm 3: Energy Aware Resource (EAR) Schedul-
Cluster Cluster name Workloads ing Algorithm) is used to schedule the resources effectively
with minimum energy consumption. RS then collects the
C1 Compute Technological computing, performance information of available resources from RID. RID contains
testing
details of all the resources available in resource pool and
C2 Storage E-com and storage and backup services
reserve resource pool. Based on cloud consumer details
C3 Communication Websites, critical internet applications, RS assigns resources and executes heterogeneous cloud
mobile computing services
workloads. Workload with highest priority is put into the
C4 Administration Endeavour software, online transaction
processing, central financial services, categories of urgent workloads and remaining will be con-
productivity applications, sidered as non-urgent workloads. SOCCER automatically
software/project development and checks the total workloads in the workload queue after each
testing and graphics oriented new workload is added. Priorities of workloads are changing
adaptively. The reason for changing priorities might be that
priority of newly added workload is higher. For this workload
Resource scheduling After successful provisioning of deadline is mandatory to consider. Otherwise, new work-
resources, resource scheduler (RS) takes the information load with higher priority waits for long time which leads to
from the appropriate workload after analyzing the hetero- starvation and reduce user satisfaction. Therefore, we used
geneous workload details which cloud consumer demanded (Algorithm 3) for this purpose.

123
1794 Cluster Comput (2016) 19:1787–1800

Resource execution During execution of a particular cloud

3.2.3 Executor (E)
workload, the Resource Executor (RE) will check the current
workload. If the resources are sufficient for execution then
Executor implements the plan after analyzing completely.
it will continue with execution otherwise request for more
To reduce the energy consumption is a main objective
resources. RE will monitor the value of energy consump-
of executor as shown in [Algorithm 4: Executing Unit
tion continually. If the value of Actual Energy Consumption
(EU)]. Based on the output given by analysis and executor
(EnCactual ) is less than the threshold value of energy con-
tracks the new workload submission and resource addi-
sumption (E t ) then RE will execute workloads otherwise
tion, and generates the alert. Effector is used to transfer the
REwill generate alert for rescheduling of resources. After
new policies, rules and alerts to other nodes with updated
successful execution of cloud workloads, RE releases the
information.
free resources to resource pool and RE is ready for execution
of new cloud workloads.

123
Cluster Comput (2016) 19:1787–1800 1795

4 Experimental setup and results and every EC has their own value of energy consumption
(kWh/EC time unit (Sec)). Table 4 shows the characteristics
We have used empirical methods to evaluate the performance of the resources used and their Execution Component (EC)
of SOCCER. Tools used for setting cloud environment for access energy consumption per time unit in kWh.
empirical evaluation are Microsoft Visual Studio 2010, Net-
Beans IDE 7.1.2, Oracle Java SDK V.6, Aneka, SQL Server
2008 and, JADE Platform (for agents). In this experimen- 4.1 Experimental results
tal setup, three different cloud platforms are used: Software
as a Service (SaaS), Platform as a Service (PaaS) and In order to validate SOCCER, we have selected two exist-
Infrastructure as a Service (IaaS). The integration of multiple ing energy efficient resource scheduling approaches i.e.
environments used to conduct experiments is shown in Fig. 4. Preemption-Aware Energy Efficient (PAEE) [9] and Energy
Cloud user interacts with SOCCER through Cloud Workload Credit Scheduler (ECS) [16] as discussed in Sect. 2. Energy
Management Portal (CWMP) to submit the workload details. consumption of SOCCER, PAEE and ECS is compared based
Note: CWMP is described in our previous research work [26]. on four different clusters of workloads [(a) Compute (C1),
At software level, Microsoft Visual Studio 2010 is used to (b) Storage (C2), (c) Communication (C3) and (d) Adminis-
provide user interface in which user can access service from tration (C4)] as shown in Table 3. With increase the number
any geographical location. At platform level, Aneka cloud of workloads, energy consumption is increasing.
application platform is used as a scalable cloud middleware Table 5 shows the different cloud workloads considered
to make interaction between IaaS and SaaS, and continually for different test cases.
monitor the performance of the system. At Infrastructure
level, three different servers (consist of virtual nodes) have 4.1.1 Test Case 1: energy consumption with different
been created through Citrix Xen Server and SQL Server has number of workloads for compute cluster (C1)
been used for data storage. Scheduler runs at IaaS level on
Citrix Xen Server. Computing nodes used in this experiment Figure 5 shows the energy consumption of different num-
work are further categorized into three categories as shown in ber of workloads (10–60) for SOCCER, PAEE and ECS in
Table 4. Energy consumption is measured in Kilo-Watt-Hour Compute Cluster. It is clearly shown that the PAEE and
(kWh) using Joule Meter. Experiment setup using 3 servers in ECS consuming more energy than SOCCER at different
which further virtual nodes (12 = 6 (Server 1) +4 (Server 2) +2 workloads. At 50 workloads, energy consumption in SOC-
(Server 3)) are created. Every virtual node has different num- CER is 8.37 % lesser than PAEE and 6.42 % lesser than
ber for Execution Components (ECs) to process user request ECS. Average energy consumption in SOCCER is 5.47

JADE
Agent SOCCER

[Algorithm 3: EAR]
Workload
Aneka Autonomic Resource Scheduler
CWMP
Workload
Executed
Analyzer
Workload
Workload Info (Clustering of Workload Execution
Workloads)
Workload
Execution Info
Resource
Manager

Resource Pool

Software as a Service Platform as a Service Infrastructure as a Service

Fig. 4 Cloud environment at Thapar University

123
1796 Cluster Comput (2016) 19:1787–1800

Table 4 Configuration details of Thapar cloud

Resource_Id Configuration Specifications Operating Number of Number Energy consumption
system virtual node of ECs (kWh/EC time unit)

R1 Intel Core 2 1 GB RAM and Windows 6 18 18

Duo-2.4 GHz 160 GB HDD
R2 Intel Core 1 GB RAM and Linux 4 12 21
i5-2310- 160 GB HDD
2.9GHz
R3 Intel XEON E 2 GB RAM and Linux 2 6 25
52407-2.2 GHz 320 GB HDD

Table 5 Cloud workloads considered for different test cases

Test Case Cluster Workload Description

Test Case-1 Compute (C1) Performance testing (Processing Larger Image File of Size 713 MB), in which
SOCCER converts an image file from JPEG format to PNG
format. Conversion of a single JPEG file into PNG is
considered as a single workload.
Test Case-2 Storage (C2) Storage and backup data Store larger amount of data (5 TB) and creates backup of data
Test Case-3 Communication (C3) Website Website of Thapar University (http://www.thapar.edu) Website
is accessed by a large number of users during Admission
Period
Test Case-4 Administration (C4) Software development Developed and tested Agri-Info (to manage agriculture related
and testing information) Software in a controlled environment. (http://
www.cloudbus.org/reports/AgriCloud2015.pdf)

Fig. 5 Comparison of energy 120

Energy Consumption (KWH)

consumption with different

number of workloads for 100
compute cluster (C1)
80

60 SOCCER

40 PAEE
ECS
20

0
10 20 30 40 50 60
Number of Workloads

and 6.66 % lesser as compared to PAEE and ECS respec- 60 workloads. SOCCER reduces 6.16–8.71 and 7.95–9.82 %
tively. average energy consumption as compared to PAEE and ECS
respectively.

4.1.2 Test Case 2: energy consumption with different

number of workloads for storage cluster (C2) 4.1.3 Test Case 3: energy consumption with different
number of workloads for communication cluster (C3)
Comparison of energy consumption for SOCCER, PAEE and
ECS in Storage Cluster with different number of workloads Figure 7 shows the energy consumption of different number
is shown in Fig. 6. SOCCER performs better than PAEE and of workloads for SOCCER, PAEE and ECS in Communi-
ECS in terms of energy consumption. SOCCER consumes cation Cluster. It is clearly shown that the PAEE and ECS
12.36 % lesser than PAEE and 14.31 % lesser than ECS at 50- consuming more energy than SOCCER at different work-

123
Cluster Comput (2016) 19:1787–1800 1797

Fig. 6 Comparison of energy 140

consumption with different
number of workloads for storage
120

Energy Consumption (KWH)

cluster (C2)

100

80
SOCCER
60 PAEE

40 ECS

20

0
10 20 30 40 50 60
Number of Workloads

Fig. 7 Comparison of energy 120

consumption with different
Energy Consumption (KWH)

number of workloads for

communication cluster (C3) 100

80

60 SOCCER
PAEE
40
ECS
20

0
10 20 30 40 50 60
Number of Workloads

loads. At 40 workloads, energy consumption in SOCCER 4.1.5 Test Case 5: energy efficiency
is 5.65 % lesser than PAEE and 7.12 % lesser than ECS.
Average energy consumption in SOCCER is 4.71 and 6.12 Energy Efficiency is a ratio of number of workloads success-
% lesser as compared to PAEE and ECS respectively. fully executed in a data center to total energy consumed to
execute those workloads. Equation (6) is used to calculate
energy efficiency.

n
4.1.4 Test Case 4: energy consumption with different Energy E f f iciencyi =
i=1
number of workloads for administration cluster (C4)  
number of workloads successfully executed in a data center
total energy consumed to execute those workloads
Comparison of energy consumption for SOCCER, PAEE
and ECS in Administration Cluster with different number of (6)
workloads is shown in Fig. 8. SOCCER performs better than
PAEE and ECS in terms of energy consumption. SOCCER In this test case, energy efficiency is measured for all the
consumes 13.76 % lesser than ECS and 17.71 % lesser than four clustered with different number of resources (Execu-
PAEE at 40–60 workloads. SOCCER reduces 7.75–9.46 and tion Components) as shown in Fig. 9. It has been depicted
10.69–12.77 % average energy consumption as compared to from Fig. 9; the value of energy efficiency is increasing with
ECS and PAEE respectively. increasing in number of resources.

123
1798 Cluster Comput (2016) 19:1787–1800

Fig. 8 Comparison of energy 140

consumption with different
number of workloads for

Energy Consumption (KWH)

administration cluster (C4)
120

100

80
SOCCER
60
PAEE
40 ECS

20

0
10 20 30 40 50 60
Number of Workloads

Fig. 9 Energy efficiency of 100

different workload clusters with
number of workloads 90
80
Energy Efficiency (%)

70
60 Cluster C1
50
Cluster C2
40
30 Cluster C3
20 Cluster C4
10
0
6 12 18 24 30 36
Number of Resources

5 Conclusions and future scope (capability to readjust resources) and self-protecting (detec-
tion and protection of cyber-attacks).
In this paper, energy efficient autonomic cloud comput-
ing system is proposed for energy efficient scheduling of Acknowledgements One of the authors, Sukhpal Singh (SRF-
Professional), acknowledges the Department of Science and Technol-
cloud computing resources in data centers. The proposed ogy (DST), Government of India, for awarding him the INSPIRE (Inno-
system (SOCCER) has been validated in real cloud envi- vation in Science Pursuit for Inspired Research) Fellowship (Registra-
ronment. SOCCER focused on energy efficient scheduling tion/IVR Number: 201400000761 [DST/INSPIRE/03/
of computing resources in virtual data centers for execution 2014/000359]) to carry out this research work. We would like to thank
all the anonymous reviewers for their valuable comments and sugges-
of heterogeneous cloud workloads and uses the autonomic tions for improving the paper.
model to optimize the energy consumption. The experimen-
tal results show that the proposed system performs better in
terms of energy consumption as compared to existing sys- References
tems. Further, SOCCER can be extended by developing a
QoS aware autonomic resource provisioning and schedul- 1. Singh, S., Chana, I.: EARTH: Energy-aware autonomic resource
ing technique which will consider self-healing (find and scheduling in cloud computing. J. Intel. Fuzzy Syst. 30(3), 1581–
1600 (2016)
react to sudden faults), self-optimization (maximize resource 2. Singh, S., Chana, I.: Resource provisioning and scheduling in
utilization and cost and time efficiency), self-configuration clouds: QoS perspective. J. Supercomput. 72(3), 926–960 (2016)

123
Cluster Comput (2016) 19:1787–1800 1799

3. Singh, S., Chana, I.: Q-aware: Quality of service based cloud 22. Polverini, M., Cianfrani, A., Ren, S., Vasilakos, A.V.: Thermal-
resource provisioning. Comput. Electr. Eng. 47, 138–160 (2015) aware scheduling of batch jobs in geographically distributed data
4. Singh, S., Chana, I.: QRSF: QoS-aware resource scheduling frame- centers. IEEE Trans. Cloud Comput. 2(1), 71–84 (2014)
work in cloud computing. J. Supercomput. 71(1), 241–292 (2015) 23. Liu, Q., Wan, J., Zhou, K.: Cloud manufacturing service system for
5. Chen, K., Hu, C., Zhang, X., Zheng, K., Chen, Y., Vasilakos, A.V.: industrial-cluster-oriented application. J. Internet Technol. 15(3),
Survey on routing in data centers: insights and future directions. 373–380 (2014)
IEEE Netw. 25(4), 6–10 (2011) 24. Mashayekhy, L., Nejad, M.M., Grosu, D., Vasilakos, A.V.: An
6. Mastelic, T., Oleksiak, A., Claussen, H., Brandic, I., Pierson, J.M., online mechanism for resource allocation and pricing in clouds.
Vasilakos, A.V.: Cloud computing: survey on energy efficiency. IEEE Trans. Comput. 65(4), 1172–1184 (2016)
ACM Comput. Surv. (CSUR) 47(2), 1–36 (2015) 25. Xu, D., Liu, X., Vasilakos, A.: V: traffic-aware resource provision-
7. Choi, S., Chung, K., Yu, H.: Fault tolerance and QoS schedul- ing for distributed clouds. IEEE Cloud Comput. 2(1), 30–39 (2015)
ing using CAN in mobile social cloud computing. Clust. Comput. 26. Singh, S., Chana, I.: Efficient cloud workload management frame-
17(3), 911–926 (2014) work. Masters Dissertation. Thapar University, India. (2013).
8. Wang, B., Qi, Z., Ma, R., Guan, H., Vasilakos, A.V.: A survey on Retrieved From: http://dspace.thapar.edu:8080/jspui/bitstream/
data center networking for cloud computing. Comput. Netw. 91, 10266/2247/1/sukhpal_singh_me_thesis.pdf
528–547 (2015)
9. Salehi, M. A., Krishna, P. R., Deepak, K. S., Buyya, R.: Preemption-
aware energy management in virtualized data centers. In: The
Proceeding of 5th IEEE International Conference on Cloud Com-
Sukhpal Singh joined Com-
puting (CLOUD), pp. 844–851 (2012)
puter Science and Engineering
10. Ren, S., He, Y., Xu, F.: Provably-efficient job scheduling for energy
Department of Thapar Univer-
and fairness in geographically distributed data centers. In: The
sity, Patiala, India, in 2016 as
Proceeding of 32nd IEEE International Conference on Distributed
Lecturer. Dr. Singh obtained the
Computing Systems (ICDCS), pp. 22–31 (2012)
Degree of Master of Engineering
11. Pelley, S., Meisner, D., Zandevakili, P., Wenisch, T.F., Underwood,
in Software Engineering from
J.: Power routing: dynamic power provisioning in the data center.
Thapar University, as well as a
ACM Sigplan Not. 45(3), 231–242 (2010)
Doctoral Degree specialization
12. Urgaonkar, R., Urgaonkar, B., Neely, M. J., Sivasubramaniam,
in “Autonomic Cloud Comput-
A.: Optimal power cost management using stored energy in data
ing” from Thapar University,. Dr.
centers. In: The Proceedings of the ACM SIGMETRICS Joint Inter-
Singh received the Gold Medal
national Conference on Measurement and Modeling of Computer
in Master of Engineering in Soft-
Systems, pp. 221–232 (2011)
ware Engineering. Dr. Singh is a
13. Shen, S., Wang, J.: Stochastic modeling and approaches for man-
DST Inspire Fellow [2013-2016]
aging energy footprints in cloud computing service. Serv. Sci. 6(1),
and worked as a SRF-Professional on DST Project, Government of
15–33 (2014)
India. He has done certifications in Cloud Computing Fundamentals,
14. Changtian, Y., Jiong, Y.: Energy-aware genetic algorithms for task
including Introduction to Cloud Computing and Aneka Platform (US
scheduling in cloud computing. In: The Proceedings of the Seventh
Patented) by ManjraSoft Pty Ltd, Australia and Certification of Rational
IEEE ChinaGrid Annual Conference, pp. 43–48 (2012)
Software Architect (RSA) by IBM India. His research interests include
15. Ma, Y., Gong, B., Sugihara, R., Gupta, R.: Energy-efficient deadline
Software Engineering, Cloud Computing, Operating System and Data-
scheduling for heterogeneous systems. J. Parallel Distrib. Comput.
bases. He has more than 30 research publications in reputed journals
72(12), 1725–1740 (2012)
and conferences.
16. Kim, N., Cho, J., Seo, E.: Energy-credit scheduler: an energy-aware
virtual machine scheduler for cloud systems. Future Gener. Com-
put. Syst. 32, 128–137 (2014)
17. Chen, C., He, B., Tang, X.: Green-aware workload scheduling in Inderveer Chana joined Com-
geographically distributed data centers. In: The Proceeding of 4th puter Science and Engineering
IEEE International Conference on Cloud Computing Technology Department of Thapar Univer-
and Science (CloudCom), pp. 82–89 (2012) sity, Patiala, India, in 1997 as
18. Wang, L., Zhang, F., Vasilakos, A.V., Hou, C., Liu, Z.: Joint virtual Lecturer and is presently serving
machine assignment and traffic engineering for green data center as Professor in the department.
networks. ACM SIGMETRICS Perform. Eval. Rev. 41(3), 107– She is Ph.D. in Computer Sci-
112 (2014) ence with specialization in Grid
19. Wang, L., Zhang, F., Zheng, K., Vasilakos, A. V., Ren, S., Liu, Z: Computing, M.E. in Software
Energy-efficient flow scheduling and routing with hard deadlines Engineering from Thapar Uni-
in data center networks. In: The Proceeding of 34th IEEE Inter- versity and B.E. in Computer
national Conference on Distributed Computing Systems (ICDCS), Science and Engineering. Her
pp. 248–257 (2014) research interests include Grid
20. Shu, Z., Wan, J., Zhang, D., Li, D.: Cloud-integrated cyber-physical and Cloud computing and other
systems for complex industrial applications. Mobile Netw. Appl. areas of interest are Software
1–14 (2015) Engineering and Software Project Management. She has more than 100
21. Wang, L., Zhang, F., Aroca, J. A., Vasilakos, A. V., Zheng, K., Hou, research publications in reputed Journals and Conferences. Under her
C, Liu, Z.: GreenDCN: a general framework for achieving energy supervision, more than 30 ME thesis and four Ph.D thesis have been
efficiency in data center networks. IEEE J. Sel. Areas Commun, awarded and four Ph.D. thesis are on-going. She is also working on
32(1), 4–15 (2014) various research projects funded by Government of India.

123
1800 Cluster Comput (2016) 19:1787–1800

Maninder Singh received his Rajkumar Buyya is a Fellow

Bachelor’s Degree from Pune of IEEE, Professor of Computer
University in 1994, and holds Science and Software Engineer-
a Master’s Degree, with hon- ing, Future Fellow of the Aus-
ors in Software Engineering from tralian Research Council, and
Thapar Institute of Engineer- Director of the Cloud Com-
ing & Technology, as well as puting and Distributed Systems
a Doctoral Degree specializa- (CLOUDS) Laboratory at the
tion in Network Security from University of Melbourne, Aus-
Thapar University. Dr. Singh is tralia. He is also serving as the
currently working as Associate founding CEO of Manjrasoft, a
Professor in Computer Science spin-off company of the Univer-
and Engineering Department at sity, commercialising its innova-
Thapar University. Dr. Singh is tions in Cloud Computing. He
on the Roll-of-honour at EC- has authored over 500 publica-
Council USA, being certified as Ethical Hacker (C—EH), Security tions and four text books. He is one of the highly cited authors in
Analyst (ECSA) and Licensed Penetration Tester (LPT). Dr. Singh computer science and software engineering worldwide (h-index 100+,
has successfully completed many consultancy projects for renowned 50000+ citations). He has served as the foundation Editor-in-Chief
national bank(s). (EiC) of IEEE Transactions on Cloud Computing and now serving as
Co-EiC of Journal of Software: Practice and Experience.

123