2017 Fourth International Conference on Software Defined Systems (SDS)

OSPF Routing Protocol Performance in Software

Defined Networks
Albert Rego1, Sandra Sendra1,2, Jose Miguel Jimenez1, Jaime Lloret1
1
Universitat Politècnica de València, Valencia, Spain
2
Dept. of Signal Theory, Telematics and Communications Department (TSTC), Universidad de Granada, Granada, Spain
[email protected], [email protected], [email protected], [email protected]

Abstract— Software Defined networks (SDNs) paradigm tries SDN introduces the concept of controller [6]. The
to improve the network performance, facilitating the network controller is the central element in a virtualized network that
management and the scalability basing on open-source software manages and controls the network devices having a centralized
and adding a new entity called controller that manages the entire view and communicating the different layers in order to
network. One of the goals of the controller is taking decisions in operate in a more efficient way. The SDN controller is the
regards to routing instead of distributing it among the network element that has to fill the routes tables.
nodes, as usual. However, this field inside SDN needs more work
because it is not clearly defined how the traditional routing In a network composed by several computers, we can
protocols will work over a SDN. Thus, in this paper, we analyze classify the different elements attending to the role these
how a dynamic routing protocol as Open Shortest Path First devices have in the transmission process. Hence, the network
(OSPF) works in a SDN. The study analyzes the stability of can be divided into three different layers:
network through parameters as convergence time and round trip
time (RTT) and the quality of service (QoS) when a video is  Application layer: it provides the different services that
streamed between end devices. The results have been compared introduce the SDN improvements.
with a traditional network in order to evaluate the similarity in
terms of stability and performance. Results show that, despite the
 Control layer: it is located in the controller itself. It has
proposal uses an OSPF implementation that does not have a the capability of determining which path has to take the
performance as good as the traditional solution in terms of traffic (a packet or a flow) and if some action should be
convergence time, with the SDN proposed we get better taken over these packets, i.e., dropping them.
performance in terms of Quality of Service.  Data layer: it forwards the traffic through the different
ports of a switch.
Keywords— software defined networks (SDNs);routing; OSPF;
multimedia content; virtualization. In SDN architecture, the control layer and the data layer are
the most important ones. The control layer must obtain data
I. INTRODUCTION from the network to build a global image of this network by
One of the most important issues in networking is the communicating with the network nodes and acting over the
problem of the efficient selection of the best path through a applications through some APIs.
network. This path should ensure the reaching of all possible
destinations and should bring the possibility of adapting new In a traditional network, these two layers are located
paths to sudden changes in network topology produced either together inside the same element, e.g., the network node. This
by links failures or by the traffic [1]. node works in a distributed way to forward the traffic or to
choose how the traffic must be treated. Thereby, routers and
Routing is a traditional problem in physical networks that switches decide about the path a packet must follow when it
can be solved by using several algorithms such as Open reaches the node itself. This decision is usually made attending
Shortest Path First (OSPF), or Routing Information Protocol criteria such as destination address or link bandwidth to
(RIP) [2] and nowadays, it is not an actual challenge. However, forward the packet through the correspondent port (data layer)
when Software Defined Networks (SDNs) appeared, the way [7][8].
of understanding the network operation radically changed.
Currently, the issue of routing in SDN should be again The initial idea of the SDN paradigm [9] was to have the
considered in order to know who these networks work. SDN is routing protocol running on the controller. However, this task
a new paradigm in the field of networking [3]. It is based on can also be located in a distributed way as independent
changing the network architecture as it is currently known and processes or applications. In this way, the routing process runs
move it to a virtual plane where network devices are virtualized in every network node that acts as a traditional router. These
and defined by commands. This new way of deploying nodes must communicate to each other and with the controller
networks offers better programmability and control over reporting to it the changes that can appear in the routes.
devices, making easier the network management and the Another important issue is to select the routing protocol or
improvement of network performance [4][5]. protocols that should be used in a SDN and what is the
controllers’ role in the routing process. The controlled should

978-1-5386-2855-3/17/$31.00 ©2017 IEEE 131

 2017 Fourth International Conference on Software Defined Systems (SDS)

decide which metric or path is the most appropriate attending latency and the scalability with the number of switches using
to the network needs. their new designed controller.
Currently, there is no information about if using traditional However, the routing problem has not been analyzed as
routing protocols like OPSF or RIP are adequate in a SDN. deeply as other topics. In fact, routing inside a SDN is one of
Moreover, it is needed to determine if an adaptation (with the the areas that needs more investigation in order to find the
minimum level of artificial intelligence) of these protocols, optimal solution and turns SDN as the technology used in the
which also communicate with the controller, may be enough to future of networking. The SDN paradigm indicates that the
provide a dynamic routing required in a SDN. controller has to build the route table and perform the routing
selection. However, the use of traditional routing protocols and
Therefore, and attending to the few available information their location in the network architecture is a problem that has
about SDN behavior in regard to dynamic routing, in this not been discussed in the literature. It is necessary a further
paper, we analyze and compare the behavior of OSPF in a research about this problem, defining the role that the
virtualized topology. Results are compared with the actual controller has to play during the routing process [13]. Hence,
topology running the same protocol. Our goal is to know if in this paper the use of a new SDN routing proposal is
there is some differences in network performance and level of described and compared with traditional networks [14]. This
quality of service (QoS) between both topologies. In this case, proposal is tested using a video streaming scenario through
OSPF is used because it is considered as the most widely used which the network performance offered by OSPF in SDN is
Interior Gateway Protocol (IGP). The virtualized topology has verified.
implemented using Mininet as emulator [10]. To perform the
tests, we transmit a streaming flow through both networks. The III. ARCHITECTURE AND TOPOLOGY
comparison has been performed in terms of convergence time
and round trip time (RTT) to measure the network stability and Before describing the topology used in the experiments, we
delay, jitter and percentage of packet loss to measure the level need to explain the architecture presented and used in this work
of QoS. in order to build our SDN with dynamic routing. For this
reason, this section is divided into two subsections. The first
This rest of this paper is structured as follows. Section II one describes the proposed SDN architecture and the software
presents some of the most relevant works related to SDN and used to implement our topology. The second subsection
routing protocols in virtualized networks. The proposed describes the topology designed as well as the hardware used in
architecture is discussed in Section III where the details about both traditional and virtual network.
how the SDN and our network are implemented are also
shown. Measurement results are shown in Section IV. Finally, A. SDN architecture
Section V presents conclusion and future work. To perform our tests, we need to define a SDN to run a
routing protocol. To do that, it is needed to define our
II. RELATED WORK architecture. The SDN runs in a virtual environment due to the
SDN has become a hot topic in the field of cost related to the expensive equipment needed to build a
telecommunications. In recent years, several published works physical SDN. Mininet is used in order to emulate the SDN.
have presented and questioned interesting aspects of SDN Mininet is a reliable emulator that runs over a Linux-based
concept relaying many different problems to solve. This operative system. So, in order to build the virtual SDN,
section shows some of this works. Mininet runs over an Ubuntu in a computer (PC).
It is possible to find works where the performance of The routing protocol runs as an independent process in
virtualized topologies developed in Mininet are analyzed. De every node which acts as a router in order to provide routing in
Oliveira et al. [11] described SDN paradigm, besides its between defined devices inside the proposed SDN, as in a real
elements and structure. They also commented several network. To implement the protocol, we have selected Quagga
interesting aspects of Mininet and how Mininet can avoid the routing suite [15]. Quagga allows implementing several routing
use of real and expensive equipment. The authors concluded protocols. However, we have focused this study on OSPF
that SDN and Mininet is needed to make easier the because of its extended use. Therefore, Quagga is executed in
implementations of networks with good performance. every node that acts as a router in the topology.
Moreover, Keti et al. [12] presented a Mininet evaluation in Quagga works emulating routers as a Linux host. This way
order to determine its limitations. The results show that the of routing emulation implies the restriction of having only one
emulation environment may introduce a significant effect in the router in a single computer. This fact forces us to split the
needed of implementing a topology. virtual SDN into several computers, one for every router in the
In [6], Shah et al, analyzed the architecture of the most topology. Furthermore, it leads to establish some
common SDN controllers (from now SDN controller are the communication links between the different physical computers
only one that uses OpenFlow as a communication standard). In that run Mininet. The communication between these computers
addition, they also analyzed the network performance with is done using Generic Routing Encapsulation (GRE) tunneling.
each of these controllers. After measuring and analyzing the Thereby, the different Mininet instances can establish
architectures, the authors defined some steps to follow in order communication to each other and forward the packets through
to increase the efficiency of a new controller design. With these the SDN. We also need to use some switches as bridges in
criteria, the authors increase the throughput and enhance the order to communicate different Mininet instances and establish
GRE ports.

132
 2017 Fourth International Conference on Software Defined Systems (SDS)

Fig. 1 shows the architecture proposed and described. Each To deploy our virtualized network, we need four different
colored zone represents a physical computer in the SDN routers. Thus, four different physical PCs (PCA, PCB1, PCB2
emulation. For instance, the yellow zone is the subnet emulated and PCC) should be used to implement Quagga routing in the
in PCA. Each physical PC in the SDN (only PCA and PCB are virtual SDN. These routers separate different subnets, i.e., from
showed) has three different software layers. Inside these subnet 20.0.0.0/8 to subnet 70.0.0.0/8. Moreover, a PC is
physical PCs, there may be several virtual PCs emulated by connected to the virtual router implemented in PCA and
Mininet. For instance, PCC has three different PCs, one for another PC is connected to the virtual router implemented in
each subnet in order to test the connectivity between subnet. PCC.
So, every physical PC runs Mininet over Ubuntu OS and
Routers from PCA to PCC are named as R1-R4. Thereby,
Quagga is executed in each virtual router defined by Mininet.
As Fig.1 shows, we have Ubuntu running as the OS. Mininet is PCs connected to R1 and R4 are used either as a server (for the
case of R1) and as a client host (for R2) in the video streaming
executed over Ubuntu and finally, Quagga runs and emulates
the routers. Moreover, Fig.1 shows that the different PCs are process. The PC connected to R1, with IP address 20.0.0.10, is
labeled as h1 and the PC connected to R4 with IP address
communicated thank to the GRE tunnels. Every packet is
forwarded to another router through the GRE tunnel 70.0.0.10 is labeled as h2. This nomenclature is used in the
next section to define which one sends the data.
established between the PCs.
On the other hand, Mininet uses Python scripts to The physical PCs used to virtualize each virtual router have
IP addresses from 192.168.0.1 to 192.168.0.4 and are directly
implement the topology. These scripts define the part of the
network that runs over the PC where it is launched. In order to connected through a LAN.
communicate these PCs through the GRE tunnels and define This topology is also built in the traditional case, so the
them, the commands shown in Fig. 2 are needed. only difference between both topologies is that in the SDN
case, we need to split the networks into four physical PCs.
IV. TEST BENCH AND RESULTS
This section presents the equipment used in our test and the
measurements results.
The network performance comparison between physical
and virtual topology is performed from two points of views. On
the one hand, we have measured the stability of network in
terms of convergence time and RTT. The second test is a real
multimedia transmission. To do it, we have selected a video
streaming which has been sent through the networks. The
values of delay, jitter and average rate of packet loss have been
measured. In the topology, only fast Ethernet links (i.e., links
that allow transmissions with a data bit rate of 100 Mbps) have
Fig. 1. Architecture proposal for every PC and communication using GRE. been used.
B. Topology In order to deploy both physical/traditional network and
virtual network, the equipment of Table 1 has been used.
With the described architecture, we can build the SDN
topology. Fig. 3 shows the topology used in this paper. TABLE I. HARDWARE USED IN BOTH TOPOLOGIES
In order to measure the parameters to define the network Hardware Used in Each Topology
stability and level of QoS, we need to have several paths to Network
Device Virtualized Topology with
reach the same destination. Thus we decided to use a ring Physical Topology
Mininet
topology since it offers the redundancy we need. Switches
CISCO Catalyst 3560 PoE
Emulated
24
Fig.3 shows how the routers are connected to the rest of Routers CISCO 2621 XM Emulated
routers through a switch. This decision has been taken because
i7- 5500U 2.4 GH i3-4160 3.60 GH 6GB
a low-speed serial link between routers (as it is commonly PCs
8 GB RAM RAM
done) would reduce the streaming . These switches work as
bridges in the virtual topology.

Fig. 2. Commands used to establish GRE tunnels

133
 2017 Fourth International Conference on Software Defined Systems (SDS)

Fig. 3. Testbed Topology..

These results show that OSPF has bigger convergence time

A. Convergence Time and RTT Tests in the virtualized network, but the network has better response
During the first test, we have tried to define the stability of due to its low RTT. Thus, virtualized networks seem to need
OSPF in both topologies. Stability is defined as the network more time to stablish the routes however it is faster in
state when there is an available path from any source to any transmission when routes are stablished.
destination. Moreover, convergence time is the quantity of time
that a routing protocol takes to reach the stable status when B. Multimedia Quality of Service Test
some links are down and some paths has been redefined to The second test tries to measure the level of QoS during a
provide services. multimedia streaming. In this test, a video is streamed from h1
to h2 in both topologies. The video has a resolution of
The convergence time of both topologies are compared in
1280x728 and has been codified with MPEG-2. Finnaly, it is
order to determine if OSPF presents similar performance in transmitted using RTP (Real Time Protocol) via VLC software
both topologies. The test consists on sending a continuous
v.2.2.4. We have measured the most common QoS parameters,
transmission from h1 to h2 through PCB1 subnet to PCB2
i.e., consumed bandwidth, transmission delay, jitter and
subnet. After that, the link is set to down and enabled again to average rate of packet loss.
reestablish the connectivity. Fig. 4 shows the convergence time
in seconds for both cases. When the network has converged, First of all, it is important to know that in the physical
i.e. the best available path to reach h2 has been defined, the topology registers bigger packet loss percentage than in the
value of stability is equal to 1. Otherwise, the value of stability virtual one. Concretely, the physical topology presents an
is equal to 0. Then, the convergence time is the difference of average value of packet loss of 58% while virtualized topology
time in seconds between the path is set down and the stability only registers an 8% of packet loss (See Fig. 6). This is caused
is returned to one. In both cases the transmission is interrupted by the capabilities of the routers and its constraints, which are
around thirteen second. As we can see, the convergence time omitted in the virtual case.
for physical topology is faster than the convergence time for
virtualized topology. The ratio of packet loss also modifies the consumption of
bandwidth during the streaming. Fig. 7 shows consumed
While physical topology takes around 10 seconds while bandwidth during the video streaming for both topologies. On
virtual topology needs about 30 seconds to recalculate the path. the one hand, the virtualized topology shows a consumed
This difference is due to the OSPF implementation in Quagga. bandwidth with peaks of 2 Mbps while the physical topology
only reaches a bandwidth of 0.4 Mbps.
The RTT is an important indicator to know the average
delay present in the topology. Fig. 5 shows the average RTT in On the other hand, we can observe that physical topology
ms. measured for both topologies. In this case, we can observe seems to have a consumed bandwidth more constant while the
that the average RTT for the virtualized topology is more than bandwidth related to the virtualized topology not only is grater
one second and a half faster than the average RTT for the but also more variant.
physical topology. The emulation reduces significantly the
delay in the network. Then, the multimedia QoS test is needed
to fully understand the real performance of our proposal.

134
 2017 Fourth International Conference on Software Defined Systems (SDS)

3
1
2.5
Physical Topology

Round Trip Time (ms)

Virtual Topology 2
Stability

1.5

0.5

0
0
Physical Topology Virtual Topology
0 20 40 60 80 100
Time(s)

Fig. 4. Stability measured in both topologies. Fig. 5. Round-Trip time for both topologies.

100%
90%
80%
70%
Packet loss (%)

60%
50%
40%
30%
20%
10%
0%
Physical Topology Virtual Topology

Fig. 6. Average value of packet loss in % for both topologies. Fig. 7. Consumed bandwidth during video streaming for both topologies.

Physical Topology Virtual Topology Physical Topology Virtual Topology

2000 200
175
1500 150
125
Jitter (ms)
Delay(ms)

1000 100
75
500 50
25
0 0
0 1000 2000 3000 4000 5000 6000 7000 0 1000 2000 3000 4000 5000 6000 7000
Sequence Number Sequence Number

Fig. 8. Delay in ms. during video streaming for both topologies. Fig. 9. Jitter in ms. during video streaming for both topologies.

The big difference between both values is caused by the streaming quality. These peaks are caused by the loss of data
lost packets discarded in some node as they pass through the bursts in the physical topology and then the next received
network. The main consequence is that the video cannot packet has a significant delay and produces a jitter increment.
properly be visualized on the destination host. Without taking into account the peaks, while the average
values of jitter remain almost similar for both topologies, the
Besides, attending to Fig. 8 where the delay, measured in
delay is slightly higher in virtualized topology and less constant
ms. during video streaming for both topologies is shown and than the delay measured in physical topology.
Fig. 9 which provides the jitter results, we can observe 2
significant peaks in delay measurements for the physical These measures prove that, although both topologies have a
network which also causes bigger jitter and a worse video similar behavior, the packet loss rate produced in the physical

135
 2017 Fourth International Conference on Software Defined Systems (SDS)

topology causes an increase in the jitter value and a REFERENCES

considerable decrement in the bandwidth consumption. [1] S. Sendra, P. A. Fernández, M. A. Quilez, J. Lloret, “Study and
Therefore, the video streaming will present less quality in the Performance of Interior Gateway IP routing Protocols”, Network
physical network than in the virtualized network. Protocols and Algorithms, 2010, Vol. 2, No. 4. Pp. 88-117
[2] V. Vetriselvan, P. R. Patil, and M.Mahendran, “Survey on the RIP,
V. CONCLUSION AND FUTURE WORK OSPF, EIGRP Routing Protocols”, Int. Journal of Computer Science and
Information Technologies, 2014, Vol. 5, No. 2, pp.1058-1065
The improvements in network virtualization and SDN are
[3] N. Feamster, J. Rexford, and E. Zegura, “The Road to SDN: An
encouraging the development of practical studies to determine intellectual history of programmable networks”, ACM QUEUE, 2013,
the behavior of this type of networks compared to the operation Vol. 11, No. 12, Pp. 1-20.
of traditional ones. [4] Jose M. Jimenez, Oscar Romero, Albert Rego, Avinash Dilendra, Jaime
Lloret, Performance Study of a Software Defined Network Emulator,
Following this idea, this paper has presented a practical test The Eleventh International Conference on Internet Monitoring and
to evaluate the behavior of OSPF in both SDN and a physical Protection. ICIMP 2016. May 22 - 26, 2016 - Valencia, Spain
topology. The virtual SDN running OSPF has been developed [5] Jose M. Jimenez, Oscar Romero, Albert Rego, Jaime Lloret, Analyzing
using Quagga and Mininet. the Performance of Software Defined Networks vs Real Networks,
International Journal on Advances in Networks and Services, vol. 9, no
The study has been performed to check the network 3&4, Pp. 107 - 116. 2016
stability and its performance when a video is streamed between [6] S. A. Shah, J. Faiz, M. Farooq, A. Shafi, and S. A. Mehdi, “An
end devices. The experiments show that OSPF presents higher Architectural Evaluation of SDN Controllers”. In proceedings of the
convergence time in virtualized topologies. However, the IEEE ICC 2013 - Next-Generation Networking Symposium, Budapest,
network presents less delay and RTT than the physical network June 9-13. 2013.
with traditional OSPF-based routing solution. [7] H. Yang, J. Zhang, Y. Zhao, Y. Ji, J. Han, Y. Lin, S. Qiu, Y. Lee, Time-
aware Software Defined Networking for OpenFlow-based Datacenter
Furthermore, regarding to the video streaming transmission, Optical Networks, Network Protocols and Algorithms, Vol 6, No 4
the SDN has some important benefits over the traditional (2014). Pp. 77-91.
network. The SDN solution has less loss rate and jitter, and a [8] E. Molina, E. Jacob, A. Astarloa, Using OpenFlow to control redundant
paths in wireless networks, Network Protocols and Algorithms, Vol 8,
better streaming quality. Moreover, the high loss rate presented No 1 (2016). Pp. 90-103
in the traditional network reduces the bandwidth consumption, [9] B. A. A. Nunes, M. Mendonca, X-N. Nguyen, K. Obraczka, and T.
especially on the host links. Hence, the proposal provides a Turletti, “A Survey of Software-Defined Networking: Past, Present, and
stable routing in a SDN, so in the future this architecture can be Future of Programmable Networks”. IEEE Communications Surveys &
used in deeper researches where dynamic routing in a SDN is Tutorials, 2014, Vol. 16, No. 3, Pp. 1617-1634.
needed. [10] Kaur, K., Singh, J., Ghumman, N. S., “Mininet as Software Defined
Networking Testing Platform”, International Conference on
As future work, we will implement bigger topologies to Communication, Computing & Systems (ICCCS.2014), 20 Feb - 21 Feb
check other routing protocols and their behavior in terms of 2014, Chennai, India
scalability. Because the use of SDN is quickly growing, we [11] R. L. S. de Oliveira, A. A. Shinoda, C. M. Schweitzer, P. L. Rodrigues,
would like to measure their power consumption [16] when "Using Mininet for Emulation and Prototyping Software-Defined
Networks", In `proceedings of the IEEE Communications and
secure methods [17] are implemented in virtual topologies. Computing (COLCOM 2014), June 4-6 , 2014, Bogota, Colombia, pp.
1-6.
ACKNOWLEDGMENT
[12] F. Keti, ans S. Askar, "Emulation of Software Defined Networks Using
This work has been supported by the “Ministerio de Mininet in Different Simulation Environments", In proceedings of the
Economía y Competitividad”, through the “Convocatoria 2014. 6th International Conference on Intelligent Systems, Modelling and
Simulation (IEEE ISMS 2015), Kuala Lumpur, Malaysia, Feb. 9-12,
Proyectos I+D - Programa Estatal de Investigación Científica y 2015. p. 205-210.
Técnica de Excelencia” in the “Subprograma Estatal de [13] T. Slattery, “Will Software Defined Networks Need Routing
Generación de Conocimiento”, project TIN2014-57991-C3-1-P Protocols?”, [On line article]. In Nojitter web site. Available at:
by the Spanish Ministry of Economy and Competitiveness and http://www.nojitter.com/post/240168080/will-software-defined-
networks-need-routing-protocols [Last access: Feb. 19, 2017].
the European Regional Development Fund (Projects TIN2013-
[14] J. M. Jimenez, O. Romero, A. Rego, A. Dilendra, J. Lloret, “Study of
46223-P, and TEC2016-76795-C6-4-R), by the “Ministerio de Multimedia Delivery over Software Defined Networks”, Network
Educación, Cultura y Deporte”, through the “Ayudas para Protocols and Algorithms, 2015, Vol. 7, No. 4. Pp. 37-62.
contratos predoctorales de Formación del Profesorado [15] P. Jakma, and D. Lamparter, " Introduction to the quagga routing suite",
Universitario FPU (Convocatoria 2015)”. Grant number IEEE Network 2014, Vol. 28, No. 2, Pp, 42-48.
FPU15/06837 and the “Programa para la Formación de [16] J. M. jimenez, O. Romero, J. Lloret, J. R. Díaz, “Energy savings
Personal Investigador – (FPI-2015-S2-884)” by the consumption on public wireless networks by sdn management”. Mobile
networks and applications, 2016. Pp. 1-11.
“Universitat Politecnica de Valencia”.
[17] Q. Yan, Q. Gong and F-A. Deng, “Detection of DDoS Attacks Against
Wireless SDN Controllers Based on the Fuzzy Synthetic Evaluation
Decision-making Model”, Ad hoc and sensor wireless networks, 2016,
Vol. 33, No.1-4. Pp. 275-299.

136