Performance Evaluation of Software Deﬁned Network Controllers

Edna Dias Canedo

1 a

, F

abio L

ucio Lopes de Mendonc¸a

, Georges Daniel Amvame Nze

Bruno J. G. Praciano

2,3 b

, Gabriel P. M. Pinheiro

and Rafael T. de Sousa Jr.

2 c

Department of Computer Science, University of Bras

ılia (UnB), Bras

ılia-DF, Brazil

Cybersecurity INCT Unit 6, Decision Technologies Laboratory — LATITUDE, Electrical Engineering Department (ENE),

Technology College, University of Bras

ılia (UnB), Bras

ılia-DF, Brazil

Department of Mechanical Engineering, University of Bras

ılia (UnB), Bras

ılia-DF, Brazil

Keywords:

Software Deﬁned Networks, Virtualization, Network Topology, Software Deﬁned Networks Controller.

Abstract:

The increasing digitization of various industrial sectors, such as automotive, transportation, urban mobility,

telecommunications, among others, reveals the need for using point-to-multipoint communications services,

such as constantly updating software and reliably delivering messages users, as well as optimizing the use

of hardware and software resources. The implementation of Software Deﬁned Networks (SDN) provides the

network with the ﬂexibility and programming capabilities needed to accurately and reliably support point-to-

multipoint distribution services. This paper presents an account of the activities implemented to produce the

performance evaluation of the SDN controllers. We identiﬁed the metrics used in the literature to perform

performance evaluation of SDN controllers. The mechanisms adopted were used to describe the performance

evaluation environment and the respective operating processes of the controllers. Besides, a methodology is

proposed to perform the performance evaluation of the controllers.

1 INTRODUCTION

Some serious problems arise in the current net-

works, such as the limitations of standardized equip-

ment that runs proprietary software, the high demand

of routing tables, the complexity of various proto-

cols, the large number of applications executed by

users which create network bottlenecks. To mitigate

these problems, a new concept of network architec-

ture, called Software Deﬁned Network (SDN) (Chen,

2019), (Karimzadeh et al., 2014) has emerged.

SDN is increasingly gaining ground in the context

of digital transformation and new technology capabil-

ities, because its resource provisioning is not much af-

fected by hardware constraints as SDN control system

is centralized allowing intelligent monitoring to adapt

the network automatically according to trafﬁc condi-

tions (Hohlfeld et al., 2018). In short, it provides a

way to scan the network (Ochoa-Aday et al., 2019)

and to consequently adapt the network services.

Therefore, the SDN design is an architecture of

networking between computers, whose main differ-

https://orcid.org/0000-0002-2159-339X

https://orcid.org/0000-0002-7423-6695

https://orcid.org/0000-0003-1101-3029

ence to traditional models is that SDN allows central-

ized control of the network through software appli-

cations. In practice, this helps the operator to man-

age the entire network more consistently, regardless

of the underlying communications technology that is

used (Ochoa-Aday et al., 2019).

The SDN paradigm breaks vertical integration by

radically separating the packet forwarding and the

control plans, providing applications with a central-

ized and abstract view of network distribution (Jung

et al., 2019). This architecture allows software de-

velopers to build new network deployment methods

tailored to different user needs to optimize the per-

formance of their applications and consequently opti-

mize their tasks (Chen, 2019). Data and control plans

are interconnected by public interfaces and allow di-

rect programming of the control plan. This way, all

network policies can be implemented and managed

logically centrally on one controller, enabling net-

work management as a single, programmable entity

(Ai et al., 2019).

This paradigm has been gaining ground and great

attention from researchers and major computer net-

work industries (Alharbi and Portmann, 2019), (Cer-

roni et al., 2019), (Umair et al., 2019), (Rasool et al.,

Canedo, E., Lopes de Mendonça, F., Nze, G., Praciano, B., Pinheiro, G. and Sousa Jr., R.

Performance Evaluation of Software Deﬁned Network Controllers.

DOI: 10.5220/0009414303630370

In Proceedings of the 10th International Conference on Cloud Computing and Services Science (CLOSER 2020), pages 363-370

ISBN: 978-989-758-424-4

363

2019), (Din et al., 2019).

The widely used and consolidated standard for

SDN is OpenFlow (Poncea et al., 2019). In SDN, all

intelligence is logically centered on software-based

controllers (control plane) and network devices be-

come simple packet forwarding elements (data plan)

that can be programmed through an open interface

provided by OpenFlow (Poncea et al., 2019).

This paper presents a proposal for a methodology

for performing the performance evaluation of SDN

controllers. The proposed methodology will be im-

plemented and used by the projects developed in the

Latitude research laboratory and will allow the per-

formance monitoring of the resources used.

The remainder of this paper is organized as fol-

lows. Section 2 presents the background. Section 3

describes the proposed model for the comparison and

performance of SDN controllers. Section 5 concludes

the paper.

2 BACKGROUND

SDN arose due to the need to automate, scale and

optimize networks, in order to better deal with ap-

plications coming from the public cloud, from pri-

vate storage services and also from databases, that

is, it was developed to follow the changes by which

service providers and operators have faced in recent

years (Alonso et al., 2019).

The SDN keep trafﬁc ﬂowing and assist with near-

instant problem resolution, offering cost savings as

costs are similar to consumption, generating long-

term savings (Gal

an-Jim

enez et al., 2018).

SDN also help improve the quality of services

and/or products offered by organizations to end con-

sumers by enabling centralized cloud monitoring,

conﬁguration, management, service delivery, control,

and automation (Schwabe, 2018).

SDN controller performs network-intensive oper-

ations, and performance is a factor that deserves at-

tention in the network role virtualization scenario, es-

pecially considering the virtualization overheads that

are already widely investigated.

In the ﬁeld of SDN controllers, the two main per-

formance metrics widely considered in various inves-

tigations are throughput and latency(Lai et al., 2019;

Li et al., 2019). Throughput refers to the performance

of the network function, namely, the amount of Open-

Flow streams the controller is capable of processing

per second, while latency measures the controller re-

sponse time to a PACKET IN message received from

a switch (Lai et al., 2019)(Li et al., 2019).

Some other secondary metrics may also be con-

sidered, such as:

• CPU Performance – A set of network functions

are sensitive to throughput and latency because

they essentially operate with packet send/receive.

However, computing is also a critical factor for the

performance of much of the virtualization of net-

work functions, so it is necessary to analyze the

percentage of system CPU utilization, particularly

in virtual environments that can impose consider-

able processing degradation and limit the overall

performance of the network function.

• Memory Performance – Managing RAM utiliza-

tion efﬁciently is a critical factor in the perfor-

mance of any computer system, especially in sce-

narios where resources are limited. Memory al-

location aspects need to be considered in per-

formance evaluation of virtualization of network

functions because overall system performance can

be negatively affected if poorly managed RAM

results in secondary (slower) memory accesses.

This situation gets even worse in virtualized sce-

narios because disk I/O is still a big performance

bottleneck for virtualized systems.

Software Deﬁned Networks is a paradigm that

breaks vertical integration with the radical separation

of packet forwarding and control plans, providing ap-

plications with an abstract centralized view of the dis-

tributed state of the network (Kantor et al., 2019).

Data and control plans are interconnected by pub-

lic interfaces and allow direct programming of the

control plan. This way, all network policies can be

implemented and managed logically centrally as a

whole on the controller, enabling network manage-

ment as a single, programmable entity. Figure 1

presents the architecture of a SDN (Bozakov, 2016).

The controller is the main element of software-

deﬁned networks because it centralizes all control and

management of the network. Thus, performance is

a critical factor in SDN and has been the subject of

much research (Bholebawa and Dalal, 2018).

In production networks, the controller manages an

OpenFlow network of real switches. However, in ex-

perimental scenarios, when implementation of a real

structure is not possible, tools that emulate switches

are generally used, such as the Cbench tool, much ex-

plored in OpenFlow controller performance investiga-

tion (Bholebawa and Dalal, 2018).

2.1 Related Works

Since the development of SDN, many comparisons

between SDN controllers have been made, mainly ad-

CLOSER 2020 - 10th International Conference on Cloud Computing and Services Science

364

Figure 1: Architecture SDN (Bozakov, 2016).

dressing a comparative analysis of Throughput and la-

tency using the Cbench (Laissaoui et al., 2015) (Khat-

tak et al., 2014).

Works typically compare various open-source

controllers, providing a quantitative analysis of

throughput and latency. Scenarios often include vary-

ing the number of switches and threads to assess con-

troller scalability (Khattak et al., 2014).

Tootoonchian et al. (Tootoonchian et al., 2012)

have pioneered comparative studies of SDN con-

trollers, the authors conducted a comparative analysis

of the effectiveness of open source controllers.

Zhao et al. (Zhao et al., 2015) conducted a com-

prehensive performance evaluation of open source

SDN controllers, also assessing how system conﬁg-

urations can interfere with controller performance.

Salman et al. (Salman et al., 2016) performed an

achievement appraisal of the latest controllers such as

ONOS and Libﬂuid-based (raw, base) controllers us-

ing Cbench (Jawaharan et al., 2018) as a testing tool

and concluded that the choice of the ideal driver de-

pends on the user’s criteria. Some work goes beyond

comparing controller performance to focus on bench-

mark tools.

Erickson (Erickson, 2013) addresses the perfor-

mance evaluation of controllers and their relationship

to the programming language used to develop them.

Alencar et al. (Alencar et al., 2014) inscribe

a comparative study between two Java-based con-

trollers (Floodlight and Beacon) from a different per-

spective: software aging. In this study, controllers

are compared primarily from a memory leak stand-

point, with Beacon outperforming Floodlight with

less memory consumption.

Rashma and Poornima (Rashma and Poornima,

2019) proposed a rapid SDN prototyping of Single

Controller (SC) and Multi Controller (MC) architec-

ture on mininet, a programming ﬂexible simulator.

The emulation displays a Multi Controller (MC) SDN

architecture which improves efﬁciency and exhibits

competence in handling scalable network on contrary

to Single-Controller (SC) architecture.

The proposed work demonstrates establishing

user-deﬁned controllers on inbuilt Open Flow con-

trollers, which also brings intra-cluster and inter-

cluster communication in a hierarchical network.

They empirically ratify the scalability of the network

by increasing the number of host nodes (Rashma and

Poornima, 2019).

de Jesus et al. (de Jesus et al., 2014) analyze re-

cent security proposals derived from the use of SDN

and verify if its usage can help the increase of con-

ﬁdence, security and privacy in cloud computing en-

vironments. Furthermore, the authors approach con-

cerns regarding security introduced by the SDN archi-

tecture and how they may compromise cloud services.

Performance Evaluation of Software Deﬁned Network Controllers

365

3 SDN CONTROLLER

PERFORMANCE

COMPARISON MODEL

This section presents the proposed environment for

the performance evaluation experiment developed

with an SDN controller implemented as a virtual net-

work function. We use the SDN Cbench controller

benchmark (Jawaharan et al., 2018), (Laissaoui et al.,

2015) to gauge the performance of the network func-

tion deployed directly on the host and on KVM and

XEN virtual machines to measure performance degra-

dation caused by the virtualization layer.

Xen is an x86 virtual machine monitor based on

a virtualization technique called paravirtualization,

which has been introduced to avoid the drawbacks of

full virtualization by presenting a virtual machine ab-

straction that is similar but not identical to the under-

lying hardware. Xen does not require changes to the

application binary interface, and hence no modiﬁca-

tions are required to guest applications (Armstrong.

and Djemame., 2011). The XEN approach is very

similar to that of KVM as is based considerably on the

works of the XEN developers. XEN supports block

devices through a hypercall that makes use of an al-

tered version of the operating system.

The VNF metrics collected in the experiments

were throughput, response time, CPU consumption

and RAM. Table 1 presents a summary of the perfor-

mance evaluation criteria of the SDN.

Table 1: SDN Performance Metrics (Lai et al., 2019; Li

et al., 2019).

ID Metric

01 Throughput

02 Latency

03 CPU performance

04 Memory Performance

To simulate a real SDN scenario, the experiments

were performed in a peer-to-peer network topol-

ogy such that a trafﬁc generator transmits OpenFlow

streams to a controller hosted on a dedicated server,

as presented in Figure 2.

The client and server are interconnected by a 1

Gbps network link. Floodlight has been set to default

according to your guidelines. Regarding the execu-

tion environment, we have used the Java language.

The client machine has two (2) processors with 12

GB of RAM and 500 GB of the hard disk. The vir-

tual machines have been conﬁgured to be identical to

the base operating system (without virtualization) that

housed the physical network function for a better bal-

ance in comparing network functions. For each ex-

periment, several tests were performed.

Figure 2 shows the SDN controller deployment

environment for performing Proposed Model Valida-

tion.

For each experiment, 30 tests were performed

with duration of 500 seconds each, using 100 MAC

addresses per switch emulated so that both the aver-

age performance for each second and the average of

the generated ﬁles were stored in text ﬁles. This sce-

nario is based on the experiments performed by Zhao

et al (Zhao et al., 2015).

The validation of the proposed model for the ex-

periments involves a set C = {1, 2, 3, 4, 5, 6} of sce-

narios, a set A of test environments that correspond to

the network functions, a set M of variations of each

environment that corresponds to the Cbench (factor)

modes and the S set of switches. Considering the sets

A = {physical function, VNF-KVM, VNF-XEN}, M

= {throughput, latency} and S = {1, 2, 4, 8, 16, 32},

the number of experiments performed to validate the

proposed model must be equal to 36.

In environment 1: physical network function, the

SDN controller must be installed directly over the op-

erating system, simulating the same architecture as a

dedicated hardware device, thus constituting a non-

virtualized architecture that we may call a native sce-

nario (Klosowski and Fiorese, 2019), as can be seen

in Figure 3 (a).

In this scenario, its necessary to remove 8 GB of

RAM from the server for a fair comparison to the vir-

tual network function in the VNF-KVM and VNF-

XEN scenarios, since the virtual machines have been

conﬁgured with 8 GB of RAM. All processor cores

were used in this scenario.

In environment 2: VNF-KVM, the network role

must be virtualized on the standard KVM, which uses

full hardware-assisted virtualization, as shown in Fig-

ure 3 (b). The emulator used must be QEMU 2.1.2.

In carrying out the experiments in this scenario, it

was necessary to replace the RAM, totaling 12 GB

in the physical machine because the VNF must have

the same amount of memory as the physical network

function (8 GB), allowing the same conﬁguration of

functions Network.

In environment 3: VNF-XEN, the network func-

tion was virtualized on the XEN 4.4 − amd64 hy-

pervisor in the default conﬁguration, namely using

paravirtualization. XEN disables automatically the

KVM hypervisor when conﬁguring XEN at system

boot, preventing KVM from consuming resources and

causing inconsistencies in performance evaluation.

Figure 3 (c) shows the VNF-XEN deployment envi-

ronment.

CLOSER 2020 - 10th International Conference on Cloud Computing and Services Science

366

Figure 2: Controller Deployment Environment.

Figure 3: VNF-XEN Deployment Environment.

3.1 SDN Performance Assessment

Process Mapping

During the performance evaluation of SDN, Cbench

was conﬁgured to emulate a speciﬁc number of

switches capable of sending streams to the controller

installed on the server for a certain period. At the end

of each performance validation, the data generated by

Cbench was collected.

Simultaneously with the execution of Cbench, the

CPU and RAM data of the physical and virtual net-

work functions were recorded by the server. It is im-

portant to note that when initiating a benchmarking,

it is important to run a script to send commands to

the server, instructing it in its network function (SDN

Controller) host CPU and memory monitoring activ-

ities. This script has stored all the data conﬁgured to

evaluate SDN performance.

The developed script should be responsible for

synchronizing the distributed system components of

the SDN benchmarking process and for emptying the

RAM, system buffers, and memory buffers at each

start of the performance monitoring process.

The script should also have the function of writ-

ing controller performance data collected by Cbench

to monitoring ﬁles on the server. Files generated dur-

ing the monitoring process should store monitoring

data on the network (controller) performance and per-

formance of physical and virtual network functions

related to system CPU and RAM utilization.

The proposed SDN performance evaluation pro-

cess for the experiments should follow the following

steps, as shown in Figure 4.

1. Deﬁne the number of switches: The performance

evaluation must be planned and the switches that

will participate in the experiment must be deﬁned

and conﬁgured, according to the proposed sce-

nario and need of the evaluation project;

2. Conﬁgure hardware and software features: All

features that will participate in the experiment

must be conﬁgured as planned in activity 1 of the

Performance Evaluation of Software Deﬁned Network Controllers

367

process;

3. Deﬁne the tools to be used during monitoring, on

both client and server: The tools should be deﬁned

according to plan and criteria adopted during the

experiment;

4. Deﬁne performance monitoring metrics: As pre-

sented in Table 1, in the literature there are pro-

posed metrics and they should be adopted and up-

dated, according to the investigated scenario;

5. Deﬁne the data collection script: In benchmark-

ing, during controller execution, a data collection

script must be designed and installed on the client

to trigger experiments and data collection.

The script authorized the start of Cbench and the

data monitoring tool that will be adopted in the

experiment. The script must be installed on the

server to send SSH commands to the server in-

structing it in its network function (SDN Con-

troller) CPU and memory monitoring activities.

The script should be responsible for synchroniz-

ing between the distributed system components of

the experiment and for emptying the RAM, sys-

tem buffers, and memory buffers at the start of a

new experiment;

6. Develop SDNs performance reports: At the end

of each experiment, reports should be generated

from the results of the data collection scripts.

3.2 Performance Indicators for SDN

Controllers

While network applications are generally sensitive to

processing, network functions perform intensive net-

work input/output operations. For example, a web

server that receives an HTTP request can send a huge

amount of CPU cycles in processing the request be-

fore sending the response to the client, while a router

basically operates with a large number of forwarding

decisions, using only few CPU cycles.

The SDN controller performs intensive network

operations, with performance being a factor that de-

serves enough attention in the VNF scenario, es-

pecially if we consider the virtualization overheads

that have already been extensively investigated. In

the area of SDN controllers, the two main perfor-

mance metrics widely considered in several scientiﬁc

researches are throughput and latency. Throughput

refers to the performance of the network function, that

is, the number of OpenFlow ﬂows that the controller

is able to process per second. Additionally, the la-

tency measures the response time of the controller to

a PACKET IN message received from a switch.

4 DISCUSSION AND THREATS

TO VALIDITY

If any performance metric of an SDN is less than 1, it

indicates that a considerable degradation of the SDN

controller’s performance occurred due to overloading

imposed by the virtualized environment. With para-

virtualization, a better result was achieved, demon-

strating the high cost of emulating network devices in

the host’s user space used during execution.

The function of a physical network engages nearly

all CPU time with the execution of the controller

and the ﬂow processing, with a reasonable number of

executions, explaining the higher performance when

compared to VNFs. The VNF-XEN utilized most

of the processing for the controller, but the process-

ing power engaged in the XEN environment, spe-

cially in privileged domain interactions, is consider-

able and results in performance degradation. When

the controller is too overloaded, the physical func-

tions as well as the virtual machines generate a great

number of interruptions. Then, the processor spends

much time idle while the VNF-XEN consumes more

processing time with the system, but employed CPU

time with ﬂow processing is longer than the remain-

ing functions of SDN processes.

Latency tests usually indicate a considerable

degradation of performance, the the additional soft-

ware layer — responsible for virtualization intro-

duced between the hardware’s host and the host’s sys-

tem — imply in higher network and data processing

complexities. This results in higher transmission de-

lay between the VNFs and its clients. The physical

function and the virtual machines consume less pro-

cessing time, as they produce more frequent interrupts

per second and leave the CPU idle for a longer inter-

val. Therefore, much of CPU time is used for virtual-

ization and system operations.

VNF-XEN works along with the privileged do-

main and strives to obtain the best possible perfor-

mance, including generating considerably less inter-

ruptions and shorter processor idle time. However,

inter-domain operations diminish CPU performance,

since XEN does not perform direct CPU scheduling

for the virtual machines, once it is not a hypervisor

integrated to the Kernel.

The hypervisor has the task to manage physical

devices of the computer, allowing for multiple in-

stances of virtual computers to be executed simultane-

ously through hardware sharing, thus optimizing re-

source usage. Virtual machine scheduling for CPU

usage is controlled by XEN instead of by the system

Performing a swapping operations during the ex-

periment could lead to lower controller performance

CLOSER 2020 - 10th International Conference on Cloud Computing and Services Science

368

Figure 4: Performance Evaluation Process SDN.

on recording ﬂow information and the state of accom-

plished connections on the hard drive, once disk I/O

operations represent a critical bottleneck in perfor-

mance of virtualized systems, posing a major chal-

lenge for current virtualization attempts.

As with any experiment, the conﬁguration of the

simulation environment may affect the obtained re-

sults, as it is known that, in some situations, the con-

ﬁgurations and used resources are prone to unforeseen

events. A way of mitigating this default is to run the

experiment in a number of environments, with differ-

ent conﬁgurations.

5 CONCLUSION

This paper presents an experiment using an SDN con-

troller implemented as virtual functions in the KVM

and XEN environments to compare the most critical

performance parameters in a native scenario to verify

the performance of network functions and the innova-

tive processes linked to the performance evaluation of

SDN. A simulation environment was built to test and

possible SDN conﬁgurations.

The experiments showed that the virtualization of

an SDN controller in the KVM and XEN environ-

ments did not overload the resources of the proposed

scenario. From the proposed scenario, we deﬁned the

mapping of the performance evaluation process to be

adopted in the experiments.

As future work, the proposed process will be

tested in various scenarios, with experiments to col-

lect metrics to verify if the adopted process is ade-

quate to evaluate the performance of SDN controllers.

These experiments will be important to validate and

evolve the proposed process.

ACKNOWLEDGMENTS

This work was supported by the Brazilian Coor-

dination for the Improvement of Higher Education

Personnel (CAPES), Grants 23038.007604/2014-69

FORTE and 88887.144009/2017-00 PROBRAL; the

Brazilian National Council for Scientiﬁc and Techno-

logical Development (CNPq), Grants 303343/2017-6,

312180/2019-5 PQ-2, BRICS2017-591 LargEWiN,

and 465741/2014-2 INCT on Cybersecurity; The

Brasilian Federal District Research Support Founda-

tion (FAP-DF), Grants 0193.001366/2016 UIoT and

0193.001365/2016 SSDDC; the LATITUDE/UnB

Laboratory (Grant 23106.099441/2016-43 SDN); the

Ministry of Economy (TEDs DIPLA 005/2016 and

ENAP 083/2016); the Institutional Security Ofﬁce

of the Presidency of the Republic (TED ABIN

002/2017); the Administrative Council for Eco-

nomic Defense (TED CADE 08700.000047/2019-

14); and the Federal Attorney General (TED AGU

697.935/2019).

REFERENCES

Ai, J., Guo, Z., Chen, H., and Cheng, G. (2019). Improv-

ing the routing security in software-deﬁned networks.

IEEE Communications Letters, 23(5):838–841.

Alencar, F., Santos, M. A., Santana, M., and Fernandes, S.

(2014). How software aging affects SDN: A view on

the controllers. In GIIS, pages 1–6. IEEE.

Alharbi, T. and Portmann, M. (2019). The (in)security of

virtualization in software deﬁned networks. IEEE Ac-

cess, 7:66584–66594.

Alonso, R. S., Sitt

on-Candanedo, I., Rodr

ıguez-Gonz

alez,

S., Garc

ıa,

O., and Prieto, J. (2019). A survey on

software-deﬁned networks and edge computing over

iot. In PAAMS (Workshops), volume 1047 of Commu-

Performance Evaluation of Software Deﬁned Network Controllers

369

nications in Computer and Information Science, pages

289–301. Springer.

Armstrong., D. and Djemame., K. (2011). Cultivating cloud

computing - a performance evaluation of virtual image

propagation & i/o paravirtualization. In Proceedings

of the 1st International Conference on Cloud Comput-

ing and Services Science - Volume 1: CLOSER,, pages

539–550. INSTICC, SciTePress.

Bholebawa, I. Z. and Dalal, U. D. (2018). Perfor-

mance analysis of sdn/openﬂow controllers: POX ver-

sus ﬂoodlight. Wireless Personal Communications,

98(2):1679–1699.

Bozakov, Z. (2016). Architectures for virtualization and

performance evaluation in software deﬁned networks.

PhD thesis, University of Hanover, Hannover, Ger-

many.

Cerroni, W., Galis, A., Shiomoto, K., and Zhani, M. F.

(2019). Telecom software, network virtualization, and

software deﬁned networks. IEEE Communications

Magazine, 57(5):88.

Chen, L. (2019). Resource Allocation in Multi-Domain

Wireless Software-Deﬁned Networks. (Allocation de

ressources dans des r

eseaux d

eﬁnis par logiciels sans-

ﬁl multi-domanes). PhD thesis, INSA Toulouse,

France.

de Jesus, W. P., da Silva, D. A., de Sousa J

unior, R. T., and

da Frota, F. V. L. (2014). Analysis of SDN contri-

butions for cloud computing security. In UCC, pages

922–927. IEEE Computer Society.

Din, S., Paul, A., and Rehman, A. (2019). 5g-enabled hi-

erarchical architecture for software-deﬁned intelligent

transportation system. Computer Networks, 150:81–

89.

Erickson, D. (2013). The beacon openﬂow controller. In

Proceedings of the second ACM SIGCOMM workshop

on Hot topics in software deﬁned networking, pages

13–18. ACM.

Gal

an-Jim

enez, J., Polverini, M., and Cianfrani, A. (2018).

Reducing the reconﬁguration cost of ﬂow tables in

energy-efﬁcient software-deﬁned networks. Com-

puter Communications, 128:95–105.

Hohlfeld, O., Kempf, J., Reisslein, M., Schmid, S., and

Shah, N. (2018). Guest editorial scalability is-

sues and solutions for software deﬁned networks.

IEEE Journal on Selected Areas in Communications,

36(12):2595–2602.

Jawaharan, R., Mohan, P. M., Das, T., and Gurusamy, M.

(2018). Empirical evaluation of SDN controllers using

mininet/wireshark and comparison with cbench. In

ICCCN, pages 1–2. IEEE.

Jung, O., Smith, P., Magin, J., and Reuter, L. (2019).

Anomaly detection in smart grids based on software

deﬁned networks. In Proceedings of the 8th Interna-

tional Conference on Smart Cities and Green ICT Sys-

tems - SMARTGREENS,, pages 157–164. INSTICC,

SciTePress.

Kantor, M., Biernacka, E., Borylo, P., Domzal, J., Ju-

rkiewicz, P., Stypinski, M., and W

ojcik, R. (2019). A

survey on multi-layer IP and optical software-deﬁned

networks. Computer Networks, 162.

Karimzadeh, M., Valtulina, L., and Karagiannis, G. (2014).

Applying sdn/openﬂow in virtualized LTE to sup-

port distributed mobility management (DMM). In

CLOSER, pages 639–644. SciTePress.

Khattak, Z. K., Awais, M., and Iqbal, A. (2014). Perfor-

mance evaluation of opendaylight SDN controller. In

ICPADS, pages 671–676. IEEE Computer Society.

Klosowski, E. A. and Fiorese, A. (2019). Freecontroller:

A framework for relative efﬁciency evaluation of

software-deﬁned networking controllers. In ICEIS (1),

pages 349–360. SciTePress.

Lai, Y., Ali, A., Hossain, M. S., and Lin, Y. (2019). Perfor-

mance modeling and analysis of TCP and UDP ﬂows

over software deﬁned networks. J. Network and Com-

puter Applications, 130:76–88.

Laissaoui, C., Idboufker, N., Elassali, R., and Baamrani,

K. E. (2015). A measurement of the response times

of various openﬂow/sdn controllers with cbench. In

AICCSA, pages 1–2. IEEE Computer Society.

Li, F., Xu, X., Yao, H., Wang, J., Jiang, C., and Guo,

S. (2019). Multi-controller resource management for

software-deﬁned wireless networks. IEEE Communi-

cations Letters, 23(3):506–509.

Ochoa-Aday, L., Cervello-Pastor, C., and Fern

andez-

Fern

andez, A. (2019). etdp: Enhanced topology dis-

covery protocol for software-deﬁned networks. IEEE

Access, 7:23471–23487.

Poncea, O. M., Pistirica, S. A., Moldoveanu, F., and Asavei,

V. (2019). Design and implementation of an openﬂow

SDN controller in NS-3 discrete-event network simu-

lator. IJHPCN, 14(1):17–29.

Rashma, B. and Poornima, G. (2019). Performance eval-

uation of multi controller software deﬁned network

architecture on mininet. In International Conference

on Remote Engineering and Virtual Instrumentation,

pages 442–455. Springer.

Rasool, R. U., Ashraf, U., Ahmed, K., Wang, H., Raﬁque,

W., and Anwar, Z. (2019). Cyberpulse: A machine

learning based link ﬂooding attack mitigation sys-

tem for software deﬁned networks. IEEE Access,

7:34885–34899.

Salman, O., Elhajj, I. H., Kayssi, A., and Chehab, A.

(2016). Sdn controllers: A comparative study. In

2016 18th Mediterranean Electrotechnical Confer-

ence (MELECON), pages 1–6. IEEE.

Schwabe, A. (2018). Data-centre trafﬁc optimisation using

software-deﬁned networks. PhD thesis, University of

Paderborn, Germany.

Tootoonchian, A., Gorbunov, S., Ganjali, Y., Casado, M.,

and Sherwood, R. (2012). On controller performance

in software-deﬁned networks. In Hot-ICE. USENIX

Association.

Umair, Z., Qureshi, U. M., Cheng, T. Y., and Jia, X. (2019).

An efﬁcient wireless control plane for software de-

ﬁned networking in data center networks. IEEE Ac-

cess, 7:58158–58167.

Zhao, Y., Iannone, L., and Riguidel, M. (2015). On the

performance of SDN controllers: A reality check. In

NFV-SDN, pages 79–85. IEEE.

CLOSER 2020 - 10th International Conference on Cloud Computing and Services Science

370