The 18th International Symposium on Communications and Information Technologies (ISCIT 2018)

Quality of Service Sensitive Routing for Software

Defined Network Using Segment Routing

Ohmmar Min Mon, Myat Thida Mon

University of Information Technology
Yangon, Myanmar
[email protected], [email protected]

Abstract—Communication networks are growing in size at an There are many flow detection techniques such as NetFlow for
increasing rate with the traditional network. Quality of Service traffic statistic polling and sFlow for packing sampling. The
(QoS) is needed for new applications and new services on the controlled switches forward packets according to the flow rules
Internet. Software Defined Network (SDN) techniques are they received.
appeared to be more flexible and programmable network. By
utilizing the characteristics of SDN, this paper proposes QoS OpenFlow managed by the Open Networking Foundation
sensitive routing for available bandwidth by using segment (ONF) is the first popular implementation of SDN [9]. The
routing (SR) to find the best path according to the QoS OpenFlow Switch checks the packet header and executes the
requirements. This paper focuses on monitoring of flow entries flow rules depending on the controller’s decision. Segment
among switches and finding the feasible path over QoS-based Routing (SR) is a new emerging traffic engineering technique
routing scheme. To provide network throughput to the end hosts, and nodes and links are assigned Segment Identifiers (SIDs)
the available bandwidth has to be accomplished. The routing [12]. The segment labels prepend with the packet to define
algorithm finds the path which is feasible to meet the desired QoS paths and the controller needs to send the segment labels to the
of data flows. If the required QoS cannot provide to the ingress router. SDN controller learns the network status
requested flow, the controller determines how to calculate information by using segments depending on the traffic path
depending on the request from the switch. If the initial path will [9]. Signaling protocols are not required to accomplish resource
not be able to achieve the available bandwidth, the algorithm
reservation. The main challenge is how to achieve the feasible
reroutes the higher bandwidth flow. This paper suggests
path for QoS flow. This paper provides available bandwidth
improvements in QoS using Open Network Operating System
(ONOS) controller with OpenFlow protocol and the simulation
measurement for the flows using OpenFlow protocol. The main
experiments with effective QoS are suitable for the future objective of this paper is to enable QoS sensitive routing in
internet using SDN. OpenFlow as one implementation of ONOS SDN.
The remainder of this paper is divided into six sections.
Keywords-Software Defined Network; Quality of service; Section II gives the background of QoS routing in SDN
Segment routing; QoS sensitive routing; Open Network Operating
environment. Section III outlines SDN and segment routing
System
architecture. Quality of Service sensitive routing is proposed in
Section IV of this paper. Section V presents the performance
I. INTRODUCTION with an evaluation experiment and test results. Finally, section
Accomplishing QoS suitability is a challenge for having VI gives the conclusions and our future research.
sufficient resources for every industry. End-to-end QoS
guarantees can give strong guarantees to end hosts. Providing II. RELATED WORK
QoS is an efficient resource management to control sufficient
Various QoS routing algorithms have been dedicated to
bandwidth, minimize packet loss and delay for numerous types
attain the best path and provide the desired performance using
of services. As the Internet develops rapidly, real-time
QoS aware routing algorithms. QoS challenges for bandwidth
applications such as voice and video conferencing increase and
and delay requirements are solved in [1]-[2]. The
users cannot acquire QoS guarantees for various services. To
accomplishment of the network by separating into bandwidth
meet the demand for QoS requirements, there is a Service
and latency adapted application for the multi-domain network
Level Agreement (SLA) between business customers and a
is discussed in [3]. They are not considered SR approach using
service provider [1]. Bandwidth is the important metric which
SDN. QoS sensitive routing is more ensuring in future Internet
concerns the QoS requirements although there are several
with SR approach.
performance constraints (e.g delay and bandwidth). Bottleneck
links will lead to congestion when bandwidth is not available QoS routing methods to measure the maximum bandwidth
for the requested flow. SDN simplifies the QoS routing and delay based on the Dijkstra algorithm between the
process. SDN Network consists of multiple SDN switches and proposed algorithm and the traditional shortest path algorithm
one controller. The SDN controller collects link status are described in [4]-[5]. Here, [4] proposed the routing function
information and calculates the traffic path in the depending on application requirements using OpenFlow while
network.

978-1-5386-8458-0/18/$31.00 ©2018 IEEE 180

Authorized licensed use limited to: UNIVERSIDADE FEDERAL DO CEARA. Downloaded on December 19,2024 at 12:42:34 UTC from IEEE Xplore. Restrictions apply.
 The 18th International Symposium on Communications and Information Technologies (ISCIT 2018)

[5] presented its task to install data paths by using link cost B. Segment Routing
values for the requested flows. From a scale and simplicity perspective, Segment Routing
The integral component of a network path is the available is especially powerful in the era of SDN with application
bandwidth for real-time applications. Most of the techniques requirements programming the network behavior. SDN
for available bandwidth measurement are divided into active controller intelligence is used to map the optimal path onto
and passive methods. Moreover [6] used a passive method to segments. Segment Routing provides proficient packet
measure available bandwidth for any time. They discussed the forwarding through the network path by specifying segment
bandwidth measurement overhead due to the passive way and list. There are SR forwarding plane actions: (a). PUSH – a
[7] solved the problem of the lack of timestamp using segment is pushed on the top of segment stack (b) NEXT –
OpenFlow. There are a large number of adaptation algorithms insert a segment to the header of SR and accomplished as an
for video streaming services over HTTP. An adaptive active segment (c). CONTINUE – forwarding the active
algorithm for DASH service [8] proposed to obtain the segment. SR procedure compares with the MPLS procedure as
appropriate path for video flows depending on the segment. It shown in Table I.
also considers the available bandwidth and bitrate of the In Segment Routing network, it is enough to have an IGP
segment. Our approach used the results for the available protocol and once Segment Routing is configured, IGP will
bandwidth of QoS flows in different network configurations take labels and redistribute them within the domain. In this
using segment routing. paper, the switch informs the controller about the requested
STE-QOSPF-Mix [10] offers the effective available flow. The controller uses the topology and determines the
bandwidth for DiffServ MPLS-TP networks. It considers the forwarding rules on the switches. The controller informs with
inconsistency between the effective available bandwidth and the OFPC_PORT_STATS messages that are provided by
the residual bandwidth. It uses a distributed approach for QoS OpenFlow protocol to the switches periodically to obtain
Routing with resource reservation. Our work does not need available bandwidth information. If a new traffic flow is
signaling protocols to accomplish resource reservation. established, the SR controller computes the path using a
segment list SID.
And [11] described the bandwidth estimation technique to
provide QoS routing using neighbor discovery procedure. It If the target path in the packet header is {103, 104, 106,
needs to discover in neighboring routers for directing traffic to 107}, node R3 uses the nodal SID 103, and sends the packet to
reach the destination. R4. R4 pops the label 103 and sends the packet with label
stack {104, 106, 107} to node R6. Node R6 knows that SID
104 has been requested at its site. So Node R6 performs the
III. SOFTWARE DEFINED NETWORK AND SEGMENT
requested service and pops the label 104 from the packet
ROUTING header. It forwards the packet with only one label {107} to
This section gives a brief explanation about SDN node R7. This process is shown in Figure 2. In this section,
framework for measuring available bandwidth using segment this system presents an example of how segment routing
routing to improve QoS. The most used standard is OpenFlow works.
as shown in Figure 1. SDN achieves an enhancement in time-
to-new features and services, network simplification, and
TABLE I. COMPARISON OF SEGMENT ROUTING AND MPLS
significant cost reduction.
Label Signaling Segment routing MPLS
A. Software Defined Networking (SDN)
Network architecture offers centralization approach that IGP/LDP SYNC Not Required Required
constructs the separation of the control function and packet FAST REROUTE
IGP IGP+RSVP-TE
forwarding. It manages the network by offering programmable FOR FRR
network services. SDN enables more flexible to obtain the SDN SUPPORT Yes No
suitable path for the requested QoS. Network intelligence in the
control plane calculates the paths for the requested flows and ROUTING TYPE Source Based Destination Based
responds the flow information to the forwarding plane. The
SCALABILITY High Low
forwarding plane forwards packets from source to destination.

Figure 1. The Architecture of Software Defined Network Figure 2. Example of Segment Routing

181
Authorized licensed use limited to: UNIVERSIDADE FEDERAL DO CEARA. Downloaded on December 19,2024 at 12:42:34 UTC from IEEE Xplore. Restrictions apply.
 The 18th International Symposium on Communications and Information Technologies (ISCIT 2018)

IV. A PROPOSED QUALITY OF SERVICE SENSITIVE ROUTING TABLE II. NOTATION LISTS

There are two approaches for QoS: approaches proposed Symbols Meaning
before SDN and SDN enabled approaches. When considering G(V,E) Network graph with vertex and edge
QoS approaches before SDN, there are two kinds of IETF
services which are IntServ services and DiffServ services. uE Utilization of edge
IntServ, per-flow design, specifies the elements to guarantee CE Edge Capacity
QoS and keep the network status information on every router.
DiffServ, a class-based architecture, operates by classifying bwu Bandwidth used by the link
the traffic into classes with per-hop behavior. However, it bwP The minimum bandwidth
cannot provide enough resources to guarantee QoS of different
flows. PS->D The set of all available paths from S to D

SDN enabled approaches to tackle all of the problems of

the traditional network. SDN is an active approach and the
controller can modify data to optimize network resources and
to simplify the routing process. Quality-of-Service is the
proficiency of a network to offer appropriate service for
different traffic. SDN controller maintains the ability to set up
forwarding rules on the network switches and to forward the
packets to the destination.

A. Some QoS requirements of applications

 Bandwidth: The maximum bandwidth to forward traffic
to the requested host without decreasing the rate of the
traffic in that path
 Hop-count: Minimum number of paths between source
and destination
 Delay: The time needed to travel across the network
that can impact on QoS from one host to another host
Figure 4. Network Topology
 Jitter: One of the quality problems that affect injury to
the quality of the packets The segmented routed path is determined by the controller
by pushing the SID to the switches. When the path is not able
This paper focuses to find the available bandwidth to acquire the available bandwidth, the QoS flow will not be
necessary of all QoS flows in SDN to meet QoS requirements. accessible. The SR controller maintains the flow for requested
In this system, the controller considers the path of the flows for information to calculate the path as shown in Figure 3. This
the switches. When a user needs a certain bandwidth, it sends a system uses Mininet simulation of the SR network topology as
bandwidth request packet to the controller. Request packet shown in Figure 4. Table II summarizes the list of
contains information such as the identification of the user, the abbreviations in this paper. Let consider network graph
destination, how much bandwidth it needs. G(V,E). In this case, C is the capacity of the edge and bwu is
the bandwidth usage which contains in the information in the
request packet. Edge utilization uE is presented in Equation.
(1).
bwu
uE  (1)
CE
For each link E, the available bandwidth resource of link E
is CE  bwu . We define the equation of the minimum
bandwidth as Equation. (2):

bwP  min (CE  bwu ) (2)

1 i  n

By using minimum bandwidth, the path that has maximum

available bandwidth between source and destination can be
Figure 3. Segment Routing Process expressed by the following formulation. (3):

182
Authorized licensed use limited to: UNIVERSIDADE FEDERAL DO CEARA. Downloaded on December 19,2024 at 12:42:34 UTC from IEEE Xplore. Restrictions apply.
 The 18th International Symposium on Communications and Information Technologies (ISCIT 2018)

bwa  max min (CE  bwu )

PPS  D 1in
(3)
Initialize# Finding Available Bandwidth
n 1
C P   C (bwu , C E )
Input: Topology: G(V,E)
Output: Available Bandwidth
i 1
(4) Bandwidth Threshold bwmax; bwf ;
Delay df, dsen ;
To get the path with the maximum available bandwidth n= number of available paths;
between source and destination, Modified Dijkstra algorithm is Pi= i-th path;
used. In this algorithm, the cost of the path, denoted Cp, is Parr1 [], Parr2 [], Parr3 []; j, k, l=0;
measured by the minimum bandwidth according to Equation Initialize: bwa=0;
Loop
(4) instead of the sum of the edge of capacities used in For (i=0, i ≤ n, i++)
traditional Dijkstra algorithm. This system is based on finding if (bwf ≥ bwmax && df ≤ dsen )
the widest path according to the bandwidth. then bwa = bwf
if 10s< dsen<20s
Flag=0; Parr1 [j] = Pi,
B. QoS Sensitive Routing
if [Pj] > [Pj-1] then select Pj;
Increasing demand for QoS sensitive traffic on the Internet else select Pj-1, j++;
grow higher and higher. The objective of our routing algorithm else if dsen<10s
is to find the path which feasible to provide the desired QoS. Flag=1; Parr2 [k] = Pi,
if [Pk] > [Pk-1] then select Pk;
Firstly, this algorithm focuses on monitoring of flow entries else select Pk-1, k++;
among switches. And then it finds the feasible bandwidth path. else Flag=2; Parr3 [l] = Pi,
If it is, higher bandwidth flow will choose to direct the flow. if [Pl] > [Pl-1] then select Pl;
else select Pl-1, l++;
To explain QoS sensitive routing algorithm, consider a endif
network topology for the network with four nodes. In the end Loop
topology network as shown in Figure 4, source node is S1 and for all n ∊ N do
destination node is S2. For bandwidth and delay sensitive flow If bwn< bwa || bwn< bwmax then bwn =0
endfor
S1-S3-S2, S1 would push SID {101,103,102}, and forward it
to S2 when the flows entered the network. For delay tolerant
flow, packets are forwarded with segment list {101,102} in the
SR header over S1 to S2, and forward it to S2. Figure 6. Algorithm of QoS Sensitive Routing

The types of flows are shown in Table III. The types of

TABLE III. TYPES OF FLOWS flows are classified into bandwidth and delay sensitive flow
(QoS flow 1), delay sensitive flow (QoS flow 2) and delay
Flow Flow’s QoS tolerant flow (Best-effort flow). Each of the flow types is
(Src, Dst) Flag
Type requirements defined according to the QoS requirements.

(H1,H3)
QoS 100 Mbps, Flag=0 (Bandwidth and In Figure 5, we used the pseudo code to emulate the
flow 1 < 20s Delay sensitive) topology of the network using Mininet, a popular network
emulator among SDN. The setup assumed for SR test uses
(H1,H3)
QoS 50 Mbps,
Flag=1(Delay sensitive) Mininet and the ONOS.
flow 2 < 10s
To create the traffic flows, ICMP requests and replies will
Best- be used. The system creates a network topology for segment
(H2,H4) Effort 110Mbps, - Flag=2(Delay tolerant) routing tests. The experiment is employed with the topology to
flow
evaluate QoS sensitive routing. Figure 6 shows the QoS
Sensitive routing algorithm. In the algorithm, bwf defines the
feasible bandwidth, bwa represents available bandwidth, df is
the feasible delay and dsen is the delay sensitive.

//Create LeafSpine Topology//

Input: Switch, Host; V. EXPERIMENT RESULTS
Initialize: S=0, H=0, N=4;
For all i=1 to N do We developed the topology with Mininet and ONOS
Si =addSwitch (Si); controller using Leaf-Spine architecture. This Leaf-Spine
Hi =addSwitch (Hi , IPi); architecture is designed to provide very scalable throughput
Li = createLink (Si); across thousands to hundreds of thousands of ports. The
Topo T = S,H,L; network topology is simulated in Mininet that constructs a set
Return T;
of virtual network interfaces to connect to the SDN controller.
The evaluation was performed on the topology depicted in
Figure 5. SR Topology Algorithm Figure 4.

183
Authorized licensed use limited to: UNIVERSIDADE FEDERAL DO CEARA. Downloaded on December 19,2024 at 12:42:34 UTC from IEEE Xplore. Restrictions apply.
 The 18th International Symposium on Communications and Information Technologies (ISCIT 2018)

TABLE IV. EXPERIMENT PARAMETERS

Parameter Values
Number of switches 4

Bandwidth threshold 30 Mbps

Delay sensitive threshold 20s, 10s

SDN controller ONOS

Simulation tool Mininet 2.2.1 Figure 8. Throughput between three test flows

Figure 7. The virtualized network in Mininet output Figure 9. End-to-end delay variaiton between three test flows

The test results show the throughput for different flows as

The script in this system is based on Python to generate shown in Figure 8. From the test results, we can see the test
flows. The network topology is comprised of 4 virtual results of QoS routing algorithm. QoS Flow 1 uses the path S1-
switches interconnected with an SDN controller, 4 virtual S3-S2 which is 80 Mbps. QoS Flow 2 uses the path S1-S4-S2
hosts and virtual Ethernet links interconnecting the switches. which is near 32 Mbps. Congestion occurs when Best-effort
The SDN controller is an Open Network Operating System flow is at 10s and it is not available to guaranteed QoS. QoS
(ONOS) controller using OpenFlow configured to flows are adequate throughput than Best-effort flow because
communicate with the data plane as shown in Table IV. QoS flows reach some peak rate and stable at 15s.
Figure 9 shows the test results with the delay variation. The
To implement SR tests, this system creates based on the QoS flows still maintained at desired seconds. Best-effort flow
ONOS project to handle the series of tests using Ubuntu increases significantly at 10s due to congestion. The delay
16.04. These are the minimum requirements to run the variation between QoS flows has decreased in 20s than Best-
environment. The PC has installed Microsoft Windows 8.1 effort flow.
OS. The ONOS controller at the network topology is used to
configure with the OpenFlow switches to emulate the routing VI. CONCLUSION
proficiencies. In this system, the controller includes a routing
policy based on a maximum bandwidth threshold 30Mbps In this paper, we have implemented available bandwidth for
desired QoS in SDN. This system presents QoS sensitive
between the switches. For this experiment, a network shown in
routing based segment routing to find the path which feasible
Figure 4 is used.
to provide the desired QoS. This system compares QoS
technique and Best-effort technique by using segment routing
The resulting output with four switches and four hosts is to provide the QoS of higher bandwidth flow. This system
shown in Figure 7. The requested services cannot provide if gives an approach to keep a good level of throughput and the
the request exceeds the threshold level of the link bandwidth. needed quality of service. The available bandwidth via
The priority of QoS flow 1 is higher than QoS flow 2 and we experiments is implemented under SDN testbed using Mininet
avoid the Best-Effort flow from using an inconsistent amount and ONOS controller. This algorithm acts perfectly using
of bandwidth at the time of congestion. Hosts h1 and h2 send bandwidth as QoS metric. Simulation results described that the
QoS traffic to h3 and h4 with the guaranteed rate of 30 Mbps. algorithm is suitable for QoS support in the future internet.
The actual rate sent by h1 is 100 Mbps and 50 Mbps Furthermore, our future work will provide packet loss
respectively. Best-effort flow rate between hosts is 110 Mbps. considerations and prove that this system is resilient to failures
All flows are generated with iperf. The received throughput is and is able to maintain the desired QoS performance.
observed from iperf’s statistics.

184
Authorized licensed use limited to: UNIVERSIDADE FEDERAL DO CEARA. Downloaded on December 19,2024 at 12:42:34 UTC from IEEE Xplore. Restrictions apply.
 The 18th International Symposium on Communications and Information Technologies (ISCIT 2018)

REFERENCES [7] P. Megyesi, A. Botta, G. Aceto, A. Pescapé and S. Molnár, “Challenges

and solution for measuring available bandwidth in software defined
networks”. Computer Communications, February 2017, pp. 48-61.
[1] L. Chienhung, W. Kuochen and D. Guocin, “A QoS-aware routing in [8] C. Cetinkaya, E, Karayer, M. Sayit and C. Hellge, “SDN for segment
SDN hybrid networks”. FNC 2017. based flow routing of DASH”. Consumer Electronics–Berlin (ICCE-
[2] S. Tomovic, I. Radusinovic and N. Prasad, “Performance comparison of Berlin), 2014 IEEE Fourth International Conference, September 2014,
QoS routing algorithms applicable to large-scale SDN networks”. 2015- pp. 74-77.
International Conference on Computer as a Tool (EUROCON), [9] D. J. Hamad, K. G. Yalda and I. T. Okumus, “Getting traffic statistics
September 2015, pp. 1-6. from network devices in an SDN environment using OpenFlow”. ITaS,
[3] H. Cho, J. Park, J. M. Gil, Y. S. Jeong, and J. H. Park, “An Optimal Path September 2015, pp. 951-956.
Computation Architecture for the Cloud-Network on Software-Defined [10] H. Zhang, Y. Zhao, D. Han, J. Zhang, J. Xing, “QoS Sensitive Routing
Networking”. Sustainability 7, no. 5, May 2015, pp. 5413-5430. in DiffServ MPLS-TP Networks”, International Conference on
[4] U. Pongsakorn, I. Kohei, U. Putchong, D. Susumu and A. Hirotake, “ Computer Application and System Modeling (ICCASM ), October 2010,
Designing of SDN-Assisted Bandwidth and Latency Aware Route Vol. 1, pp. V1-726.
Allocation”. (HPC), July 2014, pp. 1-7. [11] A. Charles and R. Bensraj, “Mobility and Bandwidth Aware QoS
[5] T. T. Nguyen and D. S. Kim, “Accumulative-load aware routing in Routing Protocol for MANET”, Research Journal of Applied Sciences,
software-defined networks”. Industrial Informatics (INDIN), 2015 IEEE Engineering and Technology, 11(1), September 2015, pp. 48-55.
13th International Conference, July 2015, pp. 516-520. [12] C. Filsfils, N.K. Nainar, C. Pignataro, J.C. Cardona, and P. Francois,
[6] P. Megyesi, A. Botta, G. Aceto, A. Pescapè and S. Molnár, “Available “The segment routing architecture”, Global Communications
bandwidth measurement in software defined networks”. Proceedings of Conference (GLOBECOM), December 2015, pp. 1-6.
the 31st Annual ACM Symposium on Applied Computing. ACM, April
2016, pp. 651-657.

185
Authorized licensed use limited to: UNIVERSIDADE FEDERAL DO CEARA. Downloaded on December 19,2024 at 12:42:34 UTC from IEEE Xplore. Restrictions apply.