A SCALABLE MONITORING SYSTEM FOR SOFTWARE DEFINED NETWORKS

International Journal of Distributed and Parallel systems (IJDPS) Vol 15, No. 1, January 2024
DOI: 10.5121/ijdps.2024.15101 1
A SCALABLE MONITORING SYSTEM FOR SOFTWARE
DEFINED NETWORKS
Mahmoud Eissa, Ahmed Yahya, Usama Gad
Department of Electrical Engineering, Al-Azhar University,
Nasr City, Cairo-11371, Egypt.
ABSTRACT
Monitoring functionality is an essential element of any network system. Traditional monitoring solutions
are mostly used for manual and infrequent network management tasks. Software-defined networks (SDN)
have emerged with enabled automatic and frequent network reconfigurations. In this paper, a scalable
monitoring system for SDN is introduced. The proposed system monitors small, medium, and large-scale
SDN. Multiple instances of the proposed monitoring system can run in parallel for monitoring many SDN
slices. The introduced monitoring system receives requests from network management applications,
collects considerable amounts of measurement data, processes them, and returns the resulting knowledge
to the network management applications. The proposed monitoring system slices the network (switches and
links) into multiple slices. The introduced monitoring system concurrently monitors applications for
various tenants, with each tenant's application running on a dedicated network slice. Each slice is
monitored by a separate copy of the proposed monitoring system. These copies operate in parallel and are
synchronized. The scalability of the monitoring system is achieved by enhancing the performance of SDN.
In this context, scalability is addressed by increasing the number of tenant applications and expanding the
size of the physical network without compromising SDN performance.
KEYWORDS
Software Defined Network; monitoring system; monitoring system architecture; network slicing;
measurements data; network management applications.
1. INTRODUCTION
One of the functions of a Network Monitoring System (NMS) is to monitor an internal network
for different types of issues or for managing the network. The NMS has the capability to identify
and assist in resolving various issues, including slow web page downloads, lost emails,
suspicious user activity, insecure network connections, and server crashes.[1].
In contrast to Intrusion Detection Systems (IDSs) or Intrusion Prevention Systems (IPSs), which
primarily focus on detecting and preventing intrusions, Network Monitoring Systems (NMSs) are
utilized to evaluate the network's performance during regular operations and gather measurement
data for network management applications, including IDSs. To effectively execute traffic
monitoring tasks, the NMS should have the capability to monitor various devices from different
vendors, ranging from mobile phones to hosts, servers, routers, and switches [2].
The SDN paradigm aims to enhance flexibility by separating the control and data planes of
conventional networks. The data plane is situated in forwarding devices enabled with SDN, such
as SDN switches, while the control plane is logically centralized in new entities referred to as
SDN controllers. This separation of planes in SDN facilitates the establishment of distinct layers

2
of abstraction, enabling a high level of flexibility in network management. Figure 1 illustrates the
architecture of an SDN-based network, showcasing the various entities, planes, and layers of
abstraction involved.[1], [3].
Figure 1. Software-Defined Networking architecture [1].
Application plane: SDN applications are granted resources by the SDN controller via Application
Programming Interfaces (APIs). These applications encompass a range of functionalities,
business applications, network management, and security management. They can be developed to
identify issues, malicious activities, anomalies, or authors. [1].
Control plane: This plane is responsible for assessing the local status of the forwarding devices
within the network and enforcing appropriate policies for the network's optimal functioning. Such
policies may include routing, traffic engineering, or security policy enforcement. Unlike
traditional networks, SDN employs flow-based forwarding decisions instead of destination-based
forwarding decisions. All packets that meet a specific criterion are handled with the same actions.
The control plane takes charge of determining the path of traffic and managing all network
signaling activities to ensure that the network devices are properly configured [1][4].
Data plane: This plane performs traffic forwarding in network devices like switches based on the
policies established by the control plane. This includes traffic filtration and various actions that
can be carried out on incoming packets in a switch. It is noteworthy that in SDN, switches are of
a general-purpose nature, implying that they implement flow-level policies established by the
control plane and can combine actions that were previously specific to different types of network
devices, such as routers and switches, in conventional networks. This plane is also referred to as
the "forwarding plane."[5].
There are numerous interfaces that exist between these layers, namely the northbound and
southbound interfaces. The northbound interface facilitates communication between the
controller and application layers, while the southbound interface enables communication between
the controller and the data forwarding layer. [6]
The SDN model is founded upon "open interfaces." The controller provides an API accessible to
control applications, enabling them to verify and change the network's status. Additionally, new
functionalities can be incorporated by developing new applications. The Controller uses open
protocols to program network nodes, such as switches and routers, in accordance with application
criteria. The most widely recognized protocol is OpenFlow, which can be deemed a standard
protocol.

3
SDN monitoring systems should possess the ability to gather vast quantities of diverse
measurement data, process them in real-time, and promptly supply the resulting knowledge to
network decision-making procedures. Meeting the demands of SDN monitoring systems is a
substantial challenge due to the current networks vast, dynamic traffic volumes, size, and the
stringent constraints on time and hardware resources, these challenges will be tackled in this
paper.
It introduces and develops a scalable monitoring system that is capable of monitoring small,
medium and large size networks. This system extends its monitoring capabilities to an increased
number of tenant applications, encompassing management applications. It monitors management
applications and allows to generate accurate and frequent monitoring reports concerning a variety
of network events and emerging conditions such as network performance bottlenecks, anomaly
detection and others. The rest of this paper is organized as follows: Section 2 shows related work
to this paper while Section 3 discusses the proposed architecture and solution. The
implementation and validation of the proposed system are discussed in section 4. Proposed
experiments for enhancing the scalability of monitoring system are discussed in section 5 while
evaluations of the proposed architecture and solution are discussed in section 6. Section 7
concludes the paper.
2. RELATED WORK
Several studies have been done to monitor Software Defined Networking. However, there are still
weaknesses in these approaches. The following section critically reviews existing research related
to this paper.
J. Suariz proposed a flow monitoring solution for OpenFlow Software-Defined Networks that
generates reports with flow-level measurements like those provided by NetFlow. To minimize the
overhead on the controller and reduce the number of flow entries needed in the switch, two traffic
sampling methods can be implemented in current SDN switches without any modifications to the
OpenFlow specifications. These methods were then implemented in the OpenDaylight controller
and their accuracy and overhead were assessed in a testbed using real-world traffic traces. [1].
B. Isyaku, M. Zahid, M. Kamat, K. A. Bakar, and F. Ghaleb presented cutting-edge research
offering various solutions to address SDN performance issues stemming from the limited flow
table. To overcome this, fuzzy theory and machine learning techniques can be leveraged to
identify frequent flow entries that should be retained in the flow table. Consequently, this paper
identifies several challenges, including update operations, resource constraints, communication
overhead, and packet processing delays [3].
M. Scarlato, on the other hand, employed three tools for network monitoring in Software-Defined
Networks to determine how these tools could be adapted to SDN. These tools include Ntop,
Wireshark, and Argus. Their performance was evaluated, and the outcome revealed that Ntop and
Argus did not behave as anticipated, as they were unable to recognize an OpenFlow traffic,
despite their capability to identify NetFlow and sFlow traffic [5].
Although Wireshark has been widely utilized as a comprehensive tool, it possesses a plethora of
filters specific to OpenFlow packets. Moreover, G. Tangary demonstrated that adaptive
monitoring functions can be integrated into the packet-processing pipeline to maintain the
accuracy of monitoring reports in the face of dynamic resource availability in the data plane [7].

4
M. Alsaeedi, M. Mohamad, and A. AL-Roubaiey have outlined four primary challenges in
OpenFlow-SDN that contribute to traffic overhead. They also highlighted research challenges
that must be addressed to develop more adaptive and scalable OpenFlow-SDN solutions under
large-scale and dynamic network conditions. One of these challenges involves monitoring and
classifying network flows at the application level [8].
3. PROPOSED MONITORING SYSTEM
In this section, the high level architecture, and detailed design of the proposed monitoring system
are introduced.
3.1. High level architecture
The high-level architecture, as depicted in Figure (2), is arranged in a layered structure where the
top layer contains management applications. Each management application has two primary
elements: an information analyzer and a decision maker. The functionality of these two elements
differs from one management application to another. For example, the functionality of the two
elements of fault detection is different from that of anomaly detection and performance
management. The information collected for each application is distinct from the others. The
decision maker receives the results of the information analyzer and makes a decision that satisfy
the objectives of the application. This decision should be sent to the routing element of the
monitor.
The monitor software layer has an information collector, a routing element, and collected
information tables. The information collector receives requests from management applications
and collects the necessary measurement data. The data needed for each application differs, so the
information collector can collect various types of data for each management application (calling
application). All collected data is stored in the collected information tables and returned to the
calling management application. The routing element receives a decision request from each
application to call the southbound interface (API Function) for reconfiguring the infrastructure
switches and links to satisfy the received decision. The detailed design of the information
collector element is shown in the figure (3).

5
Figure 2. The proposed architecture of the monitor software system
The information collector retrieves information about the infrastructure's state by sending a call to
the controller that invokes the southbound interface APIs and stores the retrieved information in
the collected information tables. The information analyzer of the management application reads
the collected information and analyzes it based on certain measurements. The analyzer's results are
sent to the decision maker, which determines the new states of the infrastructure to solve problems
such as enhancing performance. The routing element calls the controller, which in turn calls the
Southbound interface APIs to implement changes in the flows. The infrastructure layer is
simulated by a tool such as Mininet and controlled by a controller like Floodlight.

6
Figure 3: The proposed Architecture of monitor system design
The monitoring system consists of many modules: requester manipulator, translator, openflow
API caller, results collector, and results returner. The requester manipulator receives requests
from Management Applications (MAs) and stores them in the requests table. The translator
module translates each high-level request using topology tables into a low-level request and
stores it in the low-level request table. The OpenFlow API caller module sends the low-level
request to the controller, creating an OpenFlow thread to execute the required operation function
(request).
The measurement results produced by the OpenFlow threads are collected by the results collector
module. The collected results are then stored in the results table. The results returner module
associates the relevant high-level target and delivers (returns) each result to its MA request.
3.2. The Synchronization module (interface)
The synchronization interface is provided to synchronize monitoring information in a distributed
and scalable data plane. It facilitates the exchange of monitoring data between different
monitoring system instances.
3.3. The database schema of data plane topology
The topologies of data planes are represented as graphs. Each graph (data plane topology) is
represented as a set of database tables. The schema of data plane topology is represented as
follows:
Host (Host-ID, IPv4addres, IPv6address, Switch, Port)
Link (Direction, SourcePort, SourceSwitch, DestinationPort, DestinationSwitch, Type)
Switch (Switch-ID, IPv4address, Role)

7
3.4. High-Level- Requests Table schema
The requests from management applications are received by monitoring module and stored in the
high-level requests table. The schema of the high level requests table is as follows:
High-level-requests (MA-id, Req-id, task, and HL- target, Intervals).
Each high level request in the schema has the ID of the management Application (MA-id) that
sent the request followed by an identifier (Req-id). Each request has a task (needed measurements
data function), followed by the high level name of the target (node, or link). Intervals field
represents the frequencies of executing the request.
3.5. Low-level requests Table schema
This table schema is as follows:
Low-Level-requests (MA-id, Req-id, Op –code, and LL-target, Intervals).
The op-code field is the corresponding Openflow function of a task in High-Level- request.The
Low-level-target (LL-target) field is the physical address (targets) corresponding to the High-
Level -target in the High-Level -requests. Intervals field represents the frequencies of executing
the request.
3.6. Results-table schema
This schema of this table is as follows:
MA-id, Req-id, measurments-data, and timestamp.
Where MA-id is the identifier of the calling management application.The algorithm of
information collector module is shown as follows:
While there is a request from MAs // MA=Management Application
Call request manipulator ()
{
Request manipulator ( )
{
Receive a request // high level request
Store the request in HLR table //HLR=High Level Request
Call translator ()
}
Translator ( )
{
While HLR table != empty
{ Read HLR // HLR= High Level Request
Search in topology database for mapping
Map HLR to LLR //LLR=Low Level Request
Store LLR in LLR database
Call openflowAPIcaller ( )
}
}
openflowAPIcaller (
{
While LLR in low-level-request-table
{
execute openflow API equivalent to LLR
call results collector (
}
}

8
Results collector (
{
While (results
{
Receive results from openflow thread
Store results in results table
Call results returner ()
}
}
Results returner ( )
{
While (result table != empty
{
Return results to its MA caller
Call synchronization-interface ( results )
}
}
Synchronization-Interface ( information)
{
Exchange information with other monitors.
}
4. IMPLEMENTATION AND VALIDATION
For implementing and testing the proposed monitoring architecture we used the following:
1. Python programming language
2. OpenFlow:
Originally, OpenFlow defined the communication protocol in SDN architectures that allowed
the SDN controller to directly interact with network device forwarding planes, including physical
and virtual (hypervisor-based) switches and routers. This enables the network to better adapt to
changing business requirements [1].
3. Floodlight controller
The Floodlight controller is a widely used SDN controller written in Java and primarily
developed by Big Switch Networks. It is responsible for controlling and managing the SDN
network. The controller supports REST applications that can perform its functions. These REST
applications utilize Floodlight's REST API to interact with the controller, allowing them to retrieve
information from the controller and send network configurations to it. Additionally, the Floodlight
controller comes equipped with a web-based GUI [9].
4. Mininet
Mininet is a network emulator that can create a virtual network comprising hosts, switches,
controllers, and links. The hosts in Mininet run standard Linux network software, and the switches
support OpenFlow, allowing for highly customizable routing and Software-Defined Networking.
Mininet is useful for research, development, learning, prototyping, testing, debugging, and any
other tasks that could benefit from having an experimental network on a laptop or other PC [10].

9
4.1. Implementing a topology
To create the topology shown in Figure 4a, we implemented the Mininet emulator code in Figure
4b. The topology comprises four hosts, four switches, and eight links. We utilized the created
topology to collect measurement data for validating the proposed monitoring system. Figure 5
illustrates the metadata of the hosts in the created topology, while Figure 6 shows the metadata of
the topology links. Figure 7 represents the metadata of the topology switches.
Figure 4.a: The topology graph
Figure 4.b. An example for topology created using the Mininet

10
Figure 5. Metadata of the hosts of the created topology
Figure 6. Metadata of the topology links.
Figure 7. Metadata of the topology switches.

11
4.2. Test cases for validating the proposed monitoring system
In this section, two test cases as concrete examples of validating the monitoring system have been
given.
4.3. Test case1 (normal case)
In this test case, as shown in Figure 8, all links are available. The proposed monitoring system
retrieved measurement data of all links between the four switches.
Figure 8: Retrieved measurement data by Proposed monitoring system in normal case
4.4. Test case 2 (error case)
In this test case as shown in figure 9, the link between switch1 and switch2 is down (not available)
but other links are available. Therefore as shown in the Figure 9, the measurement data on the
available links retrieved by the proposed monitoring system. We notice also there is no
measurement data retrieved from the down link.
Figure 9. Retrieved measurement data by proposed monitoring system in in abnormal case
(link between swtch1 and switch2 s down).
5. ENHANCING SCALABILITY OF THE PROPOSED SDN MONITORING
SYSTEM
We address the scalability issue in different dimension enhancing the performance of SDN by
increasing TCP throughput and reducing latency supporting a large number of tenant’s
applications.

12
The scalability of proposed monitoring system and SDN achieved by slicing the physical resources
(data plane) to many slices. Each slice is controlled by a controller and monitored by a copy of the
proposed monitoring system. Large number of tenants applications can run in parallel on different
slices. Each application calls the proposed monitoring system that is installed on the target slice for
collecting measurement data needed by the application.
In this paper to achieve the scalability by slicing the physical resources, the FlowVisor has been
used. The flowvisor is a special OpenFlow controller that can slice the physical network into
multiple (potentially overlapping) virtual slices.
5.1. Slicing experiment.
We used the miniNet emulator for creating the topology shown in Figure 10. The topology has 4
hosts, 4 switches and 8 links. By using the flowvisor, this topology sliced into two slices: slice-1
(upper-slice) and slice-2 (lower-slice). To use the FlowVisor to slice the topology in Figure 10 to
two slices (upper and lower), we implemented the following code shown in Figure 11.
Figure 10. The non-sliced topology
Figure 11. The slicing code
The metadata and topologies of the two slices (slie-1 and slice-2) are shown in Figures 12.a & b,
and Figures 13.a & b respectively. The slice-1 is controlled by a Floodlight controller and
monitored by a copy of the proposed monitoring system. The slice-2 is controlled by another
copy of Floodlight controller and monitored by another copy of the monitoring system.

13
Figure (12.a). The slice-1(upper-slice) Metadata
Figure (12. b). The Slice-1 (upper Slice) graph
As shown in Figures (12.a&b) the upper slice contains three switches and two hosts. The physical
addresses of these switchesand hosts are shown in Figure (12. a).
Figure (13.a) .The slice-2 (Lower Slice) metadata

14
As shown in Figures (13.a & b) the slice-2 contains three switches. The addresses of the switches
are shown in the lower slice metadata (Figure 13.b).
5.2. Throughput and latency of non-sliced (scenario-1) and sliced (scenario-2)
topologies
We have conducted two scenarios to compare between non-sliced topology performance and the
two sliced topologies (higher and lower) performance. The performance metrics are TCP
throughput and latency.
Figure (13.b). The slice-2 (Lower Slice) Graph
In the first scenario the non-slicing topology is managed by only one floodlight controller and
monitored by one copy of the proposed monitoring system and many TCP-packets have been sent
between switches.
As shown in Figure 14, the non-slice topology is controlled by a Floodlight controller and
monitored by a copy of the proposed monitoring system that collects measurement data to tenant
applications.

15
Figure 14.Tenant’s applications run on the non-sliced topology
In the second scenario, two slices have been created and each slice is controlled by a separate
floodlight controller and monitored by a copy of the proposed monitoring system. Each
monitoring copy synchronizes with each other for data exchange
Figure 15. Tenant’s applications run in parallel on the slices using different
copies of the monitoring system.

16
As shown in Figure 15, each slice is controlled by a Floodlight controller and monitored by a
copy of the proposed monitor where the two monitoring copies are synchronized. Each
monitoring copy collects measurement data to tenant applications. These applications run on the
two slices in parallel.
The performance of scenario-1 and scenario-2 has been measured by measuring the throughput
and latency metrics. The results of measuring the throughput of each scenario are shown in
Figure.16 (a,b,c,d) and results of measuring the latency of each scenario are shown in Figures 17
(a,b,c,d)
Figure.16.a: Switch2 to switch1 TCP throughput without slicing
Figure.16.b: Switch2 to switch1 TCP throughput slice 1 (upper slice)
From Figure 16.a and Figure 16.b, we notice that the throughput of slice-1 (upper slice) is greater
than the throughput of non-slicing topology. This is because the overhead of the slice controller is
reduced due to the reduction of traffic collision. The same thing the throughput of slice2 (lower
slice) is greater than the throughput of non-sliing topology as shown in Figures 16.c and 16.d

17
Figure.16.c: Switch3 to switch1 TCP throughput without slicing
Figure.16.d: Switch3 to switch1 TCP throughput of slice2
In addition to increasing the throughput, more than one tenant applications run in parallel on the
generated slices (slice-1 and slice-2 in scenario-2) as shown in Figure 15, which verifies multi-
tenants applications.
The latency of scenario-1 (based non-sliced topology) and latency of scenario-2 (based on the two
slices) have been measured. The results are shown in Figures 17 (a, b,c, and d).In the following
figures, RTT stands for Round Trip Time.
Figure 17.a Latency of s2 to s1 in non-sliced

18
Figure 17.b Latency of s2 to s1 in slice-1
In Figure 17.a, and 17.b we can see that the latency between the two switches S2 and S1 in
scenario-1 (non-sliced topology-based) is greater than the value of latency between the same two
switches in slice-1 (scenario-2). The maximum value of latency in Scenario-1 (non-sliced
topology-based) nearly equal 0.04ms but in slice-1 (scenario-2) is nearly equal 0.0275ms (i.e. less
than 0.03ms). This means the latency (factor of performance) of the slice-1(scenario-1) is better
than the latency of scenario-1(non-sliced topology-based).
Also, in Figures 17.c, and 17.d, we notice that the latency of slice 2 (scenario-2) between switch 3
and switch 1 is between 0.04ms and 0.045ms but the latency of scenario-1 (non-sliced topology-
based) between switch 3 and switch 1 is nearly equals 0.07ms. This means that the latency of
slice-2 (scenario-2) is better than the latency of scenario-1(non-sliced topology-based).
Figure 17.c Latency of s3 to s1 in non-sliced
Figure 17.d Latency of s3 to s1 in slice-2

19
Since the throughput of each slice (scenario-2) is better than the throughput of scenario-1 (non-
sliced topology-based) and the latency of each slice (scenario-2) also better than the latency of
scenario-1(non-sliced topology –based), then the performance of scenario-2 (slices-based) is better
than the scenario-1 (non-sliced network-based). Also the number of tenant’s applications that run
in parallel on the slices of scenario -2 are increased and monitored by copies of the synchronized
proposed monitoring system.
From the above discussion, the proposed monitoring system is scalable in two dimensions (i)
increasing the size of data-plane without sacrificing the performance. This is by slicing the data-
plane where each slice monitored by a copy of the proposed monitoring system where all copies
are synchronized together. (ii) Increasing the number of tenant applications that run in parallel
and monitored by the proposed monitor. By experiments that have been conducted in this research
we noticed that the throughput and latency of each slice is better than the non-sliced. Thus, in
addition to the above two dimensions the performance (throughput and latency of SDN) of slicing
scenario is better than non-sliced scenario.
6. PERFORMANCE EVALUATIONS
We have implemented and tested the proposed scalable monitoring system, which can retrieve
measurement data from the simulated data plane based on the management application request.
The monitor can receive a request from a management application to collect measurement data,
which may differ from one application to another.
The proposed monitoring system is scalable, capable of managing large-scale networks by
dividing them into multiple slices. Each slice is managed by a copy of the proposed monitoring
system, and all copies are synchronized for exchanging information through the synchronization
module of the monitoring system. Also, the proposed monitoring system monitors multitenant
applications. The proposed scalable monitoring system has high throughput and low latency than
the non-slicing based monitoring system. The limitation of the proposed monitoring system is
working only on Floodlight SDN controller. In the future, it will be enhanced to work with other
SDN controllers.
Based on the common criteria between our proposed monitoring system solution and other solutions
we compared between our proposed scalable monitoring solution and Vivi Monita solution [11] in
table -1. Also we compared our proposed monitoring solution with Tangari's Research in table-2.
In the first comparative study, we conducted a side-by-side examination of our research and the
notable work conducted by Vivi Monita, as outlined in the 2023 paper titled "Network Slicing Using
FlowVisor for Enforcement of Bandwidth Isolation in SDN Virtual Networks." [9]. Our study
concentrates on creating a scalable monitoring system for Software-Defined Networks (SDNs),
strategically employing FlowVisor to facilitate network slicing and enhance scalability without
compromising performance. Conversely, Vivi Monita's research is centered on achieving tenant
isolation and enforcing bandwidth isolation in SDN virtual networks through the adept utilization of
FlowVisor. By closely analyzing the methodologies and outcomes of each study, we aim to detect the
distinctive contributions and methodologies employed in the dynamic field of our proposed solution
and Vivi Monita's Study [11]

20
Table 1: Acomparative study between our proposed monitoring solution and Vivi Monita solution
Aspect Our proposed monitoring solution Vivi Monita's Study [11]
Aims of
Study
A scalable monitoring system for SDNs.
Ensuring tenant isolation in SDN using
FlowVisor and an SDN controller.
Methodology
Utilized FlowVisor for effective network
slicing
Two types of renters and testing
procedures were used for connectivity and
functionality.
Key findings
Scalable monitoring system for various SDN
scales.
Correct host linkage achieved without
turning on FlowVisor.
Multitenancy
approach
Multiple copies run in parallel. Hosts can only communicate within the
same tenant.
Contribution
Addressing scalability concerns in SDNs. Correct host linkage and tenant isolation
through network slicing.
In the second comparison (Table 2), we compared our research on scalable SDN monitoring with
Tangari's [7] work on resource-efficient monitoring for future networks. Our emphasis lies in
scalability without compromising performance, while Tangari introduces innovative
methodologies for accurate monitoring. We assess key aspects, including scalability models,
monitoring frameworks, adaptive solutions, and performance metrics, aiming to highlight
distinctive contributions and advancements in both studies, offering insights into the evolving
landscape of SDN monitoring
Table 2: A comparative study between our proposed monitoring solution and Tangari's solution
Aspect The proposed scalable monitoring system Tangari's Research [7]
Objective Scalable SDN monitoring system. Accurate and resource-efficient
monitoring for future networks.
Scalability Increasing the number of tenant applications
and the size of the physical network without
sacrificing SDN performance.
Increasing network sizes, avoiding
processing bottlenecks in large-scale
networks.
Monitoring
Framework
Slices the network using FlowVisor tool,
synchronized copies using OpenFlow protocol.
Decentralized and self-adaptive
monitoring framework (SAM) with
modular architecture for diverse
management applications.
Adaptive
Solutions
Synchronized copies running in parallel,
adaptive solution for efficient data extraction
(SAM).
SAM for dynamic reconfiguration under
dynamic traffic conditions, MONA for
resilience to bottlenecks in software
dataplanes.
Measurement
Technique
Processing measurement data from different
slices in parallel.
Classifiers used to reduce measurement
data-processing costs for various
monitoring queries.
Performance
Metrics
Performing scalability without sacrificing SDN
performance. High throughput and low latency
Enhanced monitoring accuracy levels for
various measurement tasks.
7. CONCLUSION
In this paper, a scalable monitoring system for SDN is proposed.The scalability has been
addressed in two dimensions without sacrificing the performance of SDN. The two dimensions
are increasing the size of SDN and increasing the number of tenants and their applications.
The monitoring system is a vendor-independent, capable of retrieving measurement data from
any infrastructure device, including switches, routers, links, and others. This is because the

21
monitoring system uses the OpenFlow protocol for retrieving measurement data, which is
infrastructure-independent. The scalability has been achieved by slicing the topology of SDN to
slices using flowvisor tool. Each slice has been controlled by a floodlight controller and
monitored by a copy of the proposed monitoring system. MutliTenant applications run in parallel
on multi-slices.
The proposed scalable monitoring system has been tested and evaluated. The results of evaluation
show that the number of tenants’ applications can be increased without sacrificing the
performance. Also, the size of the network can be increased without sacrificing the performance
by slicing the networks to slices monitored by the proposed monitoring system. The scalable
monitoring system is working only with Floodlight SDN controller.In the future the proposed
monitoring system will be enhanced to be adaptable (i.e. it works with more than one SDN
controller).
Statements and Declarations
• Funding: no funding was received to assist with the preparation of this manuscript.
• Financial interests: The authors declare they have no financial interests.
The authors have no competing interests to declare that are relevant to the content of this
article.
• Consent to participate: not applicable
• Conflict of interest: The authors declare they have no Conflict of interest.
REFERENCES
[1] Jose Su ´ arez-Varela, Pere Barlet-Ros, Flow monitoring in Software- ´Defined Networks, Computer
Networks (2018), doi:10.1016/j.comnet.2018.02.020.
[2] Andres J. Aparcana-Tasayco ; Fredy Mendoza-Cardenas ; Daniel Diaz-Ataucuri “Open and Interactive
NMS for Network Monitoring in Software Defined Networks” 2022 International Conference,
Publisher: IEEE.
[3] Isyaku, B., Zahid, M.S., Kamat, M., Bakar, K.A., & Ghaleb, F.A. (2020). Software Defined Networking
Flow Table Management of OpenFlow Switches Performance and Security Challenges: A Survey.
Future Internet, 12, 147.
[4] C. Lee, C. Yoon, S. Shin, and S. K. Cha, “INDAGO : A New Framework For Detecting Malicious SDN
Applications,” 2018 IEEE 26th
Int. Conf. Netw. Protoc. pp. 220–230, 2018, doi:
10.1109/ICNP.2018.00031.
[5] Scarlato, Michele. (2014). Network Monitoring in Software Defined Network, Master on New
Technologies in Computer Science. Itinerary Networks and Telematics, Università degli studi di
Cagliari Cagliari, Sardinia, Italy
[6] Y. Liu, B. Zhao, P. Zhao, P. Fan, and H. Liu, “A survey: Typical security issues of software-defined
networking,” China Commun., vol. 16, no. 7, pp. 13–31, 2019, doi: 10.23919/j.cc.2019.07.002.
[7] Tangari, Gioacchino; (2019) Accurate and Resource-Efficient Monitoring for Future Networks.
Doctoral thesis (Ph.D), UCL (University College London).
[8] Alsaeedi, M., Mohamad, M. M., & Al-Roubaiey, A. A. (2019). Toward Adaptive and Scalable
OpenFlow-SDN Flow Control: A Survey. IEEE Access, 7, 107346–107379..
[9] Hasan ÖZER 1*, İbrahim Taner OKUMUŞ “A Scalable and Efficient Port-Based Adaptive Resource
Monitoring Approach in Software Defined Networks” The Journal of Graduate School of Natural and
Applied Sciences of Mehmet Akif Ersoy University 13(1): 9-26 (2022).
[10]http://pakiti.com/sdn-101-mininet-openflow-and-
gns3/#:~:text=Mininet%20is%20a%20network%20emulator,routing%20and%20Software%2DDefined
%20NetworkingM. Young, The Technical Writer’s Handbook. Mill Valley, CA: University Science,
1989.

22
[11]Monita, V., Wendato, W.& Anggiratih, E. “Network Slicing Using FlowVisor for Enforcement of
Bandwidth Isolation in SDN Virtual Networks “.Jurnal Ilmiah Teknik Elektro Komputer dan
Informatika, Vol. 9, No. 3, September 2023.
AUTHORS
Mahmoud Eissa, a dedicated master's student in Electrical Engineering at Al Azhar University,
Cairo, Egypt. He earned his Bachelor's from Future University, Egypt, with a 'good' grade in
2015 specializing in Electrical and Communication. His outstanding graduation project,
"Performance Evaluation of Routing Protocols for Mobile Ad-hoc Network," received an
excellent grade, showcasing his commitment to advancing the field.
Ahmed Yahya [ORCID: 0000-0002-3271-058X], is currently working as a Professor in the
Department of Electrical Engineering, A-Azhar University, Cairo, Egypt. He obtained his Ph.D.
from Ain Shams University, Cairo, Egypt. He has published many research papers, which
contributed significantly to the development of his field. He had been working as As Professor in
various reputed Institution’s in Abroad and Egypt. His research area focuses on Distributed Cloud
Edge Computing, 5G Testbeds, LoRaWAN Networks, V2X Communications, Reconfigurable
System-on-chip Design, Metamaterial-based Reconfigurable Antenna and Intelligent Cognitive
Radio Networks.
Usama Gad, an Assistant Professor in the Department of Electrical Engineering, Electronics and
Electrical Communications program, Faculty of Engineering, Al Azhar University, cairo, Egypt. He
obtained his Ph.D. in Electronics and Communications in 2014, focusing on "Image Processing using
Residue Number System."he earned an M.Sc. in Electronics and Communications (2008), exploring
"Realization of digital filters using residue number system," and a B.Sc. with honors (2002) for his
project on "Optical Voice Link."

A SCALABLE MONITORING SYSTEM FOR SOFTWARE DEFINED NETWORKS

Recommended

More Related Content

Similar to A SCALABLE MONITORING SYSTEM FOR SOFTWARE DEFINED NETWORKS (20)

Recently uploaded (20)

A SCALABLE MONITORING SYSTEM FOR SOFTWARE DEFINED NETWORKS