International journal of Computer Networking and Communication (IJCNAC)Vol. 1, No.

2(November 2013)

A Comparative Analysis of the Performance of VoIP Traffic with Different Types of Scheduling Algorithms
Damilola Osikoya1, KasinathBasu2
Department of Computing ,Oxford Brookes University,Oxford, UK
[email protected]

Department of Computing ,Oxford Brookes University, Oxford, UK

Abstract
The key QoS parameters for VoIP are delay, jitter and loss. In the Internet, VoIP requires the underlying packet switched network to minimize the impact of these parameters. A major contributing factor in this regard is traffic engineering carried out by scheduling algorithms. This paper studies the behavior of different types of scheduling algorithms on the delay, jitter and loss QoS parameters. The performance evaluation involves identifying the scheduling algorithms which are most suitable for VoIP communications. The result from the analysis also shows the impact of the QoS parameters on VoIP over the Internet.

Keywords:QoS Parameters, VoIP, Scheduling Algorithm, Internet 1 INTRODUCTION

Voice over IP (VoIP) is a telecommunication service that uses packet switched network infrastructure such as the Internet to facilitate interactive voice communication using telephone or computer, [8]. This approach is significantly different from traditional telecommunication service which is based on a circuit switched infrastructure. The later approach is more reliable, but significantly more expensive and wasteful in terms of resources. VoIP in contrast uses existing packet switched networks to support VOIP calls, [7]. However, since packet switched network was initially designed mainly to support traditional data traffic, it lacks any inherent support to facilitate the real-time and interactive nature of the voice calls. Therefore, traffic engineering of the existing packet switched network is essential to support VoIP. To a user, QoS in voice over IP is an attempt to get the best possible quality of voice in a call. It is a measure of the quality of service delivered to a user, [7]. The performance of a network can be characterized by a set of parameters called Quality of service (QoS) parameter. Some of the key QoS parameters for VoIP include delay, jitter, loss and error. It is important that a network can support these QoS requirement in order to successfully provide a VoIP service. One of the main challenges of a network that affects the QoS is congestion. Congestion is caused due to limited resources such as link bandwidth, processor capacity and buffer space. This results in bottleneck in the forwarding devices such as routers resulting in delay, jitter and loss.
www.arpublication.org

Congestion can be handled by appropriate admission control policy along with optimizing the usage of various resources within a network. One significant area of optimization is the scheduling algorithms used within the routers. The use of appropriate scheduling algorithm based on the nature of traffic can reduce congestion and achieve significant improvement in the QoS performance of the network. There have been several proposed scheduling schemes for supporting the QoS requirements of VoIP traffic over a packet switched network. Some of the most prominent algorithms include Weighted Fair Queuing (WFQ), Priority Queuing (PQ) and Custom Queuing (CQ). This paper presents a comparative analysis of the performance of these scheduling algorithms on the main QoS parameters of VoIP traffic. The paper is organized as follows: Section 2 gives a brief insight on QoS mechanism for VoIP networks; Section 3 provides the literature review; Section 4 presents the brief discussion on traffic characteristics of VoIP focusing primarily on the key QoS parameters; Section 5 describes the simulation set-up and VoIP traffic attributes; Section 6 presents an analysis of the simulation results; and finally Section 7 presents the conclusion of this research.

QOS MECHANISM FOR VoIP

Different mechanisms provide QoS for IP networks. Because the Internet connects multiple domain such as Autonomous systems, the integration between two domain is very important in achieving proper end-to-end QoS Delivery [14]. Although there are different frameworks for implementing QoS, the two main architectures are; Integrated Services (IntServ) and Differentiated Services (DiffServ).

2.1

IntServ

Integrated service (IntServ) is an architecture developed by IETF and is based on per flow resource reservation. It requires that an application must first make reservations before it is able to transmit to the network. To make this reservation, the Internet reservation protocol RSVP is required. IETF RFC2210 describes both the IntServ architecture and the RSVP application. To reserve bandwidth in a network, RSVP over a system where the user requests QoS for each session, . In a voice session, the SIP client sends an RSVP path message to the receiver. Each node along the path on receipt, verifies the message by checking its resources before sending, [3]. As the message reaches the user, a message is sent back to the sender through the network.

2.2

DiffServ

DiffServ is an internetwork architecture described in IETF RF C2474 [10] and updated in RF C3260 [14]. DiffServ provides QoS, [13], by enforcing policies in a network to provide Service Level Agreements in the network. It also uses a stateless approach to reduce the use of nodes in a network [10]. DiffServ is a per-aggregate-class service that uses packet tagging, [3]. It also uses Type of Service (ToS) flag in the IPv4 header which matches with the flag in the IPv6 ag. DiffServ classifies network classes and gives each packet a separate treatment based on the settings of the ToS flag. It provides a Bandwidth Broker (BB) for resource allocation and also makes sure that network resources are evenly provisioned [2].
www.arpublication.org

International journal of Computer Networking and Communication (IJCNAC)Vol. 1, No. 2(November 2013)

2.3

MPLS

MPLS offers a very good potential for integrating quality of service. It is used to support voice services in a VoIP network, [3]. MPLS can be used to establish label switching paths between ingress and egress points, thereby creating a tunnel for labeling traffic. Transmitting traffic over a label switched path guarantees it`s delivery as long as the allocated bandwidth is not exceeded. When a packet ingresses into the network through a label switching router (LSR), the router determines the label switching path (LSP) to use as it adds a label to the path. When a packet is forwarded, the intending router receives it and decides on the outgoing interface and label value based on the incoming interface and label value, [3].

RELATED WORKS

This paper looks into how different scheduling algorithms behave and their effect on delay, jitter and loss QoS parameters. The performance evaluation involves identifying the scheduling of algorithms which are most suitable for VoIP communications. We have used OPNET to perform the experiments and analyzed the results based on the impact of the QoS parameters on VoIP over the Internet. There are several related works that have also being carried out by other researchers in this field including [6], [9], [11], [4]. While [6] and [11] focused on the analysis of VoIP parameters over voice traffic for transport protocol support and efficiency, [9] focuses on the factors a affecting the growth of VoIP. As discussed by [3] the problems associated with QoS for VoIP networks they propose a framework that provides a base on which the quality VoIP networks can be built using an MSF approach. This approach is designed to assist vendors and operators in deploying QoS capable VoIP networks. Other authors, [2] provided a detailed insight into implementing QoS frameworks such as IntServ over DiffServ networks for providing QoS on VoIP applications. This work is a simulation based experiment. The simulation tool used for the experiment is OPNET network simulator.

TRAFFIC CHARACTARISTICS IN VOIP

VoIP exhibits certain properties that affect the quality of voice delivery over a network. These properties primarily include delay, jitter and loss, [12]. Delay is the time taken for a packet to travel from source to destination. QoS is assured in VoIP when delay is less than or equal to 150ms. However, it is still acceptable if delay is between 150ms and 400ms. Delay can be reduced by using appropriate codec and queuing algorithms. It can also be reduced by sending fragmented packets over the network and reassembling them at the destination. This however increases the overhead load on the network device. Jitter in VoIP is a variation in delay. It is the difference between the minimum and the maximum end-to-end delay and signifies the variable delay within the network. Packets transmitted at constant rate are expected to be received at constant rate. However, due to network conditions packets may arrive at variable rate. To balance jitter, a jitter-buffer is used to tailor out all packets received at irregular intervals and shape them to constant intervals before the packets are processed.
www.arpublication.org

Apart from delay and jitter, packet loss and error also affects the quality of voice delivery. Packet loss is primarily due to buffer over flow within the router`s memory but could also be as a result of bad transmission, late delivery or general network errors. Unlike in data applications where lost or errored information can be retransmitted, in VoIP retransmission delay is unacceptable and therefore traffic engineering policies has to be implemented to limit loss to a minimal level. In this context, scheduling algorithms have an important role to prioritize and forward the real time traffic over traditional data traffic. Other mechanism such as resource reservation, admission control and active queue management could also reduce the level of packet loss.

SIMULATION SETUP

The simulation was setup using the OPNET network simulator. The simulation environment consists of network components communicating with 2 Ethernet gateways located on two office floors which act as routers. The first floor of the office consists of two VoIP client workstations, one FTP client and one video client. The second floor of the office consists of four workstations, three of which act as VoIP clients

Figure 1: VoIP and Video over an Office Network and the fourth act as video server and one FTP server. Both routers are connected to each other via the PPP DS1 cable. The PPP DS1 connects two systems running IP and it has a data rate of 1.544Mbps. Other devices are connected to the routers with the 10 Base T cables which transmit data at 10Mbps and have a maximum length of 100 meters. The model includes an application definition, a QoS definition and a problem definition. The application definition consists of a list of the applications to be run on the network such as VoIP applications, FTP applications, and video applications. The applications send traffic into the network which are tested and analyzed. The QoS definition characterizes protocols that are supported at the IP layer. It defines different scheduling
www.arpublication.org

International journal of Computer Networking and Communication (IJCNAC)Vol. 1, No. 2(November 2013)

algorithms such as FIFO, WFQ, Custom algorithm and PQ. The pro le configuration is used to create user profiles. It can be specified on different network interfaces to create traffic on the application layer. Since this simulation focuses on Quality of Service in VoIP, hence the QoS profile has been included in the simulation and different scheduling algorithms have been set for the study 5.1 VoIP Application Attribute

The experiments are based on G.711 and G.729 encoder. The G.711 encoder is based on uncompressed Pulse Code Modulation (PCM) voice and could be used to encode straight from a traditional telephone network, [5]. The G.711 use 8 bits per sample and each sample is generated every 125 microseconds with the use of PCM and this leads to a bit rate of 64 Kbps. The G.729 generates compressed voice and typically operates at 8kbps and is ideal for VoIP because of its low bandwidth. Voice speech consists of talk spurt length and silence length which have default values. Talk spurt is defined in VoIP as the length of sound in-between silence period. The Type of Service (ToS) field value is set to interactive voice. The G.711 voice frame used in OPNET is 32 bytes long and therefore of 4 milliseconds duration. The first set of experiments from Section 6.1 to 6.4 is based on using one G.711 frame per packet whereas Section 6.5 is based on 20 frames per packet. Section 6.6 describe G.729 encoder scheme with 1 and 80 voice frames per packet.

SIMULATION RESULTS

The simulations result demonstrates the impact of scheduling algorithms on the main QoS parameters (delay, jitter and loss) of VoIP. Although these algorithms have their own advantages and disadvantages, the focus of the experiments has had an impact on the quality and performance of VoIP traffic. Different types of VoIP traffic have been analyzed with FIFO, PQ, WFQ and CQ scheduling algorithms. Each of the experiments was run for ten minutes of simulation time and the following parameters were observed: IP packet loss, point to point delay, end-to-end delay, and delay variation. The G.711 encoder scheme was used as a standard for the first set of experiments. At the next stage another set of experiments were performed using the G. 729 encoder. In each case, we compared the effect of the scheduling algorithms on the QoS parameters.

Table 1: G.711 Encoder Scheme

6.1

Packet Loss

Figure 2: Point to point queuing delay and Dropped traffic (packets/second) Figure 2 shows the performance of the four different scheduling algorithms during a period of congestion at the router. As shown in the figure above (Fig. 2), both FIFO and CQ suffers sever packet loss after around2minutes into the simulation as packets begin to get queued up in the buffer. The drop rate increases with time with CQ showing worse performance than FIFO. In the case of PQ and WFQ there is no loss till about six minutes and after which there is occasional loss in rare instances. Therefore, it can be concluded that if packets enter the network according to their assigned weights and priority as in the case of PQ and WFQ, then congestion and loss can be significantly minimized. 6.2 Point to Point Queuing Delay

Figure 3 shows the point-to-point queuing delay between floor 1 and floor 2 routers. The two routers are connected with the PPP DS1 with a data rate of 1.544 Mbps. In all the cases, delay is not significant

Figure 3: Point to point queuing delay

International journal of Computer Networking and Communication (IJCNAC)Vol. 1, No. 2(November 2013)

and is constant in the range 0.00250 to 0.00270 between the period 2 and 10 minutes of simulation time. Therefore, the scheduling algorithms have minimal effect on the point-to-point delay of the network 6.3 End-to-End Delay

Figure 4: Voice packet end-to end delay Figure 4 shows the effect of different queuing algorithms on the end-to-end delay. For FIFO, delay increases exponentially from about 1 minute and 40 seconds into the simulation after which it remains constant at 1.28 seconds. This means that as the queue builds up, the buffer experiences delay. Since packets are queued according to their order of arrival and FIFO offers best e ort service, delay is experienced when incoming VoIP traffic is held back by other types of traffic. There may also be increase in delay when a particular application hogs the whole buffer preventing flow of traffic. The red dotted line shows the graph for PQ. Using the PQ algorithm, packets are scheduled according to their assigned priority. This reduces the effect of delay in the buffer as VoIP packets are prioritized over other data packets. The above graph shows that delay is constant for PQ just a little above zero mark. Custom Queuing behavior was identical to priority queuing with packets experiencing minimal delay. After the simulation, no result was generated for WFQ which shows that with WFQ the VoIP packets did not experience any delay 6.4 Delay variation

Figure 5 shows the impact of the various queuing algorithms on the packet to packet delay variation. In all the cases, initially there is no jitter till 1 minute and 40 seconds into the experiment as the queuing
www.arpublication.org

Figure 5: Voice Packet Delay Variation buffers are empty. For WFQ and CQ jitter stays at zero throughout the experiments which shows that there is no significant delay variation with both the algorithms. For PQ, there is a minimal jitter thereafter but its stays constant throughout the remaining part of the experiment. In the case of FIFO queue the VoIP are treated on a best-e ort basis and therefore jitter becomes very prominent after the initial period. The result shows an exponential increase thereafter which however falls and stabilizes after a period. 6.5 Analysis with the G.711 Encoder Scheme

The above sets of experiments were based on using one voice frame per packet. This section presents the results of rerunning the same experiments with 20 frames per packet.

Figure 6: Delay Variation

International journal of Computer Networking and Communication (IJCNAC)Vol. 1, No. 2(November 2013)

Figure 6 shows the effect of jitter on the four different queuing algorithms with G.711 encoder scheme and 20 voice frames per packet. In G.711 with1 voice frame per packet the worst case jitter value for FIFO was 0.16 second; here the same jitter has risen to 1.20 signifying that an increase in the number of voice frames per packet would also increase the jitter in the network. This is due to the fact that more packets would have to be processed. Priority queuing also experiences a small amount of jitter, but custom queuing and WFQ shows no significant delay variation according to the graph. In the case of end-to-end delay for G.711 encoding scheme, the delay for the FIFO algorithm rises exponentially and then stabilizes. This is partly identical to the trend in section 4.3 with 1 frame per packet. The delay for CQ, WFQ and PQ remains negligible.

Figure 7: End to End Delay

6.6

Analysis with the G.729 Encoder Scheme

These set of experiments shows the point-to-point delay between the first floor and second floor routers using G.729 encoder with 1 and 80 voice frames per packet.

www.arpublication.org

10

Figure 8: G.729 with 1 voice frame per packet

Figure 9: G.729 with 80 frames per packet Figure 8 and 9 shows the point to point delay with 1 and 80 frames per packet. There is less delay in the G.729 with 1 voice frame per packet compared to 80 frames per packet. In the former case the delay is 0.004 seconds whereas in the latter case it is 0.0065.Therefore, the point-to-point delay increases between the first floor router and second floor router when the number of frames per packet increases.

7 CONCLUSION
VoIP is rapidly replacing the traditional PSTN. However, even with its rapid growth, flaws such as congestion which results in dropped packet, delay and jitter and a affects the quality of the voice communication needs to be resolved before VoIP can fully take over the telephony world. This paper presented an analysis of the queuing algorithms on the main QoS parameters of VoIP traffic. From the analysis, it is evident that WFQ, PQ and CQ offer better QoS to VoIP. These scheduling algorithms significantly minimize delay, jitter and loss of the VoIP traffic.
www.arpublication.org

International journal of Computer Networking and Communication (IJCNAC)Vol. 1, No. 2(November 2013)

11

The paper also demonstrates that the performance of the scheduling algorithms is not affected by the nature of the VoIP codec viz. G.711 and G.729. Similarly, the number of frames per packet has minimal effect on the WFQ, PQ and CQ algorithms. It is however evident from the experiments that FIFO is not suitable for VoIP and is very sensitive to the overall traffic load in the network during period of congestion. It is therefore recommended that priority queuing, weighted fair queuing and custom queuing should be used for handling voice traffic over the Internet.

REFERNCES

[1] A. Amin. VoIP Performance Measurement Using QoS Parameters. The Second International Conference on Innovations in Information Technology), 2005. [2] R. El-Haddadeh, G. A. Taylor, S. J. Watts, et al. Towards scalable end-to-end QoS provision for VoIP applications. 2004. [3] C. Gallon. Quality of Service for Next Generation Voice Over IP Networks. Technical report, 2003. [4] J.-S. Han, S.-J. Ahn, and J.-W. Chung. Study of delay patterns of weighted voice tra c of end-to-end users on the VoIP network. International Journal of Network Management, 12(5), September/October 2002. [5] M. Karam and F. Tobagi. Analysis of the delay and jitter of voice tra c over the Internet. INFOCOM 2001. Twentieth Annual Joint Conference of the IEEE Computer and Communications Societies. Proceedings. IEEE, 2:824{833, 2001. [6] M. J. Karam and F. A. Tobagi. Analysis of the Delay and Jitter of Voice Tra c Over the Internet. Twentieth Annual Joint Conference of the IEEE Computer and Communications Societies, pages 824{833, 2001. [7] B. Keepence. Quality Of Service for Voice Over IP. colloquium on Service over the internet, pages 4/1{4, 1999. [8] J. Kim, I. Lee, and S. Noh. VoIP QoS(Quality of Service) Design of Measurement Management Process Model, April 2010. [9] E. McPhillips. The Factors Affecting the Growth of VoIP, 1999. [10] A. C. Odinma and L. Oborkhale. Quality of Service Mechanisms and Challenges for IP Networks. The Pacific Journal of Science and Technology, 7(1), 2006. [11] . Papadimitriou and V. Tsaoussidis. End-to-end Support for Multimedia QoS in the Internet. ch016, 2009. [12] O. Spaniol and M. Gunes. Voice over IP. RWTH Aachen University, 2004. [13] Z. Wang. Internet QoS: Architectures and Mechanisms for Quality of Service. 2001. [14] W. Zhao, D. Olshefski, H. Schulzrinne, et al. Internet Quality of Service: an Overview. Technical report, 2000.

www.arpublication.org