2009 IEEE Symposium on Industrial Electronics and Applications (ISIEA 2009), October 4-6, 2009, Kuala Lumpur, Malaysia
Video Quality of Service in DiffServ-Aware
Multiprotocol Label Switching Network
*J. Jaffar, **H. Hashim, **H. Zainol Abidin, **M.K. Hamzah *Universiti Kuala Lumpur, 1016, Jalan Sultan Ismail, 50250 Kuala Lumpur, Malaysia **Faculty of Electrical Engineering, Universiti Teknologi MARA, 40450 Shah Alam, Malaysia ibahhashim@gmail.com
Abstract The current trend set by service providers of enhancing the degree of provisions in IP networks while maintaining the optimum quality, has promoted many research activities. These activities include those which aim to introduce improved quality of services for popular Internet services and applications. In this paper, quality of services (QoS) of video transmission on Differentiated Services (DiffServ) with multiprotocol label switching (MPLS) network is being simulated and evaluated by a network Simulator, OPNET. The need of integration of MPLS with DiffServ particularly for the video traffic has been demonstrated in the simulation. The objective of this work is to study the influence of the QoS mechanism via DiffServ-MPLS on network parameters such as packet loss, delay and throughput for different video resolutions. The comprehensive study showed general improvement in the throughput and packet loss particularly of video transmission when using DiffServ-aware MPLS network as compared to when only MPLS or DiffServ is employed. KeywordsVideo; QoS I. INTRODUCTION Various multimedia applications such as video streaming, VOIP and video conference are gaining demand bringing with it a massive congestion to the IP networks. With the emergences of multimedia applications in IP networks, bandwidth consumption has become a critical issue among the Internet community and Internet providers alike. Since bandwidth is a scarce and precious resource in the Internet, various studies of QoS mechanism had been done to satisfy the QoS of the multimedia applications to the end user. The fundamental objective of any QoS mechanism is to ensure that excessive congestion does not occur for the packets with assured QoS. QoS does not create capacity, but only support prioritization of traffic and allocation of capacity under congested conditions, or reduced the source rates to decrease congestion [2]. Therefore, the fact that the transmission of multimedia traffic needs a stable network bandwidth due to its sensitivity to packet delay, jitter and packet loss, the Internet Engineering Task Force (IETF) had introduced several QoS mechanism, including Differentiated Services (DiffServ) in order to provide the necessary QoS [1]. DiffServ is capable of providing a stateless network which is scalable and robust. The Differentiated Services architecture is designed to provide differing level of QoS to different traffic flows [2]. Following the DiffServ advantages, service providers have benefited the latter to roll out revenues by defining DiffServ architecture into their network. By assigning applications to different classes of service according to the Service Level Agreements (SLAs), guaranteed quality of services can be met. However, DiffServ alone is not enough to guarantees SLAs. Hence, IETF had implemented MPLS architectures which use labels to forward the packets into the appropriate path. The combine use of the Differentiated Services (DiffServ) and Multiprotocol Label Switching (MPLS) technologies is envisioned to provide guaranteed quality of service (QoS) [2]. In general, the bottleneck of transmission is in routers. In conventional routing, each packet is forwarded hop by hop using routing table based on algorithm in each router. Each packet is forwarded independently in every router. This approach is unreasonable and inefficient for multimedia traffic [3]. Therefore, MPLS architectures were introduced to utilize labels to forward the packets. In many implementations, DiffServ is implemented embedded with MPLS due to MPLS capability to support DiffServ architecture (Metering, marking, policing, queuing and scheduling). MPLS is being used as a forwarding scheme while DiffServ as a QoS mechanism. In the coming sections, MPLS and DiffServ architectures and its operational behavior will be briefly discussed and followed by a discussion on the interoperability of these two architectures in order to meet the quality of service (QoS) requirements. In this paper, the QoS performance evaluation will be conducted by sending an increasing streaming video onto DiffServ-aware MPLS network by utilizing the OPNET network modeler. In the simulation section, the procedures of the experiment will be explained and the results obtained will be further evaluated and discussed in the results section. QoS performance parameters involved in this experiment are the throughput, delay and packet loss. II. MULTIPROTOCOL LABEL SWITCHING (MPLS) OVERVIEW Multiprotocol Label Switching (MPLS) protocol was proposed by IETF in 1997. MPLS was introduced to improve the scalability of network-layer routing, provide routing flexibility, increase network performance, and simplify the integration of equipment using non-IP forwarding paradigms. MPLS is a packet-forwarding technology which uses labels to make data forwarding decisions. Routers that support the MPLS protocol suite 978-1-4244-4683-4/09/$25.00 2009 IEEE 963 2009 IEEE Symposium on Industrial Electronics and Applications (ISIEA 2009), October 4-6, 2009, Kuala Lumpur, Malaysia are known as Label Switching Routers (LSRs). LSRs make forwarding decisions based on a label added at the shim header between link-layer and network-layer headers. Since MPLS operates between layer 2 and layer 3, routing process will be much faster. An MPLS label has 32 bits divided among the label value (20 bits), a Time To Live (TTL; eight bits), a CoS (three bits), and a Bottom of Stack (BoS; one bit). The LSR uses the label value to make forwarding decisions; the ingress LSR sets the TTL, a counter that is decremented by each LSR along the path. If the TTL expires, the LSR discards the packet. Finally, the CoS field indicates class of service or drop precedence. Label Edge Router (LER) is an edge LSR that makes the boundary of the MPLS domain. Ingress LER will perform a PUSH operation which tags a packet with an MPLS label and enters the MPLS network; an egress LER will then do the POP operation that removes the MPLS label from a packet as it leaves the MPLS network. SWAP operation is performed by the LSRs in the MPLS domain to swap the labels according to the destined path which is called Label Switch Path (LSP). In MPLS, IP packets can be assigned to a different Forward Equivalent Class (FEC). In FEC, IP packets can be classified according to the packets priority which will determine how the packets are handled within the network. Thus all packets which belong to the same FEC get treated in the same way and get quickly routed along their path. Figure 1 show how an IP packet goes into an MPLS network while MPLS tag switching is done on each end of LER and LSR for non-edge MPLS router. Therefore, instead of using the conventional Internet packet forwarding scheme, where complicated address matching is performed at each hop in the network, MPLS forwarding has made the process far easier and faster.
Figure 1: MPLS Domain III. DIFFSERV OVERVIEW Differentiated Services or DiffServ is a routing specification standardized by the IETF that specifies a more scalable solution for classifying, managing network traffic and providing Quality of Service (QoS) guarantees on modern IP networks. Differentiated Services allow packets that require higher QoS to be classified as higher priority than other traffics. For example, video streaming can receive higher QoS than other data traffic such as Http by using DiffServ. The differentiated services architecture is based on a simple model where traffic entering a network is classified and possibly conditioned at the boundaries of the network, and assigned to different behavior aggregates. Each behavior aggregate is identified by a single DS code point. Within the core of the network, packets are forwarded according to the per-hop-behavior (PHB) associated with the DS code point [5]. DiffServ aggregates traffic into different classes by marking IP packets, which will receive specific Per-Hop- Behavior (PHB) at every node. The DiffServ Code Point (DSCP) in the Differentiated Services (DS) field of the IP header identifies the Per Hop Behavior (PHB) associated with the packet, which is used to specify queuing, scheduling, and drop precedence [5]. In a DiffServ environment, there are two defined per hop behaviors (PHB): expedited forwarding (EF), and assured forwarding (AF). EF PHB supports low loss, low delay, and low jitter. AF PHB defines four relative classes of service with each class supporting three levels of drop precedence. IV. SIMULATION MODEL All the simulations in the paper are performed on a Network Simulator, OPNET. Figure 1 show the network topology used in this experimental study. The scenarios used are based on fixed bandwidth which divides the bandwidth along the classes as follows: 75 % for Video Traffic (EF) 15% for FTP Traffic (AF21) 10% for Http Traffic (AF11) Using fixed bandwidth allocation facilitates identifying the influence of Diffserv-MPLS architecture on video quality of service. The core links were configured as E1 links (2.048 Mbps) while the links between the clients/server to the edge routers were 10Mbps. The E1 links should provide ample bandwidth for three simultaneous 512Kbit/sec video conferencing sessions, depending on the amount of other traffic on the network. The routing protocol was OSPF; therefore the best effort path would be from Router A-D-E and hence would constitute the bottleneck. There are four different scenarios being tested which are 1) Baseline Best effort topology, 2) DiffServ topology, 3) MPLS topology and 4) Integration of MPLS and DiffServ (DiffServ-MPLS). TABLE I VIDEO TRAFFIC DEFINITION Video Traffic Frame Interval (frames/sec) Frame Size (pixels) Low Resolution 10 128x120 High Resolution 15 128x240
These scenarios were tested by increasing the video traffic model as in Table 1 to observe and compare the efficiency of these QoS architecture in addressing IP packet loss, video throughput and video end-to-end delay besides observing the FTP and Http traffic flows in the network simulation. The FTP and Http traffic are obtained from OPNETs standard application traffic model for nodes which can be defined as heavy or light. In this case, FTP traffic was set to high load and best effort type of service, where files are 50000 bytes and time between client request is exponentially distributed with mean 360 seconds. For Http traffic, heavy browsing was selected 964 2009 IEEE Symposium on Industrial Electronics and Applications (ISIEA 2009), October 4-6, 2009, Kuala Lumpur, Malaysia where, page inter-arrival times are exponentially distributed with mean 60 seconds, and each page has 1000 bytes of text and 5 medium images. In the simulation, three classes of service are provided: Expedited Forwarded (EF), Assured Forwarded namely AF11 and AF21. The EF traffic has an optimum bandwidth guarantee with low latency, low jitter and no packet loss. Video traffic was defined as the EF service in the simulation. However, due to the nature of the video traffic having some burst period, with the guaranteed bandwidth, there is still some packet loss happened [2]. In the simulation, 75% of the link bandwidth was guaranteed for EF traffic. AF class should have a minimum bandwidth guarantee with low packet loss. AF21 was defined for FTP traffic while F11 was configured for Http traffic. Here, Http traffic had received the least QoS guarantees. In the simulation, background traffic has been specified at the core links. The primary purpose of the background traffic is to model the effect of general traffic in the network on selected traffic of interest. This effect primarily [9] takes the form of delay, utilization and throughput of the link which enables modeling the traffic flow as in real network. In the simulations, the background traffic was configured as 2Mbps loads in order to create enough traffic to make the link congested. Router A and Router E were the Ingress and Egress routers, while Router B, Router C and Router D were the LSR routers. For MPLS configuration, LSP1 took place at Router A - Router D - Router E and LSP2 was Router A - Router B - Router C - Router E.
Figure 2: Network Topology
AF11 and AF21 will be routed to LSP2 and class EF will route to LSP1. Hence, these traffics will avoid the bottleneck at the shortest link and offers better guaranteed services to video traffic. Table 2 showed the configured QoS parameters as to define the DiffServ contribution to traffic QoS. In DiffServ-MPLS protocol, EF class used priority queuing while AF11 and AF21 used Weighted Fair Queuing (WFQ) in each passing node. Policing using Committed Access Rate (CAR) was used at the Ingress to police the packets flow accordingly before going into the MPLS/DiffServ domain. CAR is a policing algorithm that enables the service provider to control the flow rate from one link to another link that has different bit-rates requirement. Random Early Drop (RED) was used as the congestion avoidance algorithm to randomly drop packets at each node as incoming congestions are detected. Each QoS attributes were configured at the interfaces of core routers. The advantage of employing DiffServ-MPLS in the IP network is the capability of the service provider to make full use of Forward Equivalent Class (FEC) by DiffServ traffic classification via PHB. FEC is [4] a group of IP packets which are forwarded in the same manner (e.g. over the same path, with the same forwarding treatment). TABLE II QOS PARAMETERS Router DiffServ Only DiffServ-MPLS A IF8: Traffic inbound policing (CAR) , WFQ, RED IF3: Traffic outbound policing (CAR) , WFQ, RED IF8: Traffic inbound policing (CAR), WFQ, RED IF3: Traffic policing (CAR), Priority Queuing, RED B IF1: WFQ, RED IF1: WFQ, RED C IF1: WFQ, RED IF1: WFQ, RED D IF1: WFQ, RED IF1: Priority Queuing, RED E IF7: WFQ, RED IF3: WFQ, RED IF7: WFQ, RED IF3: Priority Queuing, RED V. SIMULATION RESULTS AND ANALYSIS The simulations had demonstrated pure IP network only provide best effort services for FTP, Http and video traffic flows. All the traffic used the shortest path (A-D-E) and exceeded its bandwidth capacity, while the longer path (A-B-C-E) was under-utilized. The throughput increased at the links as the video traffic increased its traffic rate from low resolution to high resolution. Packets get dropped and delayed as buffers overflow because the resources in the network cannot meet all traffic demand. In DiffServ-MPLS, the network path is fully utilized since video traffic followed the A-D-E path while FTP and Http is routed using A-B-C-E path. The overall performance of the network had proved that DiffServ-MPLS approach had guaranteed a lower delay and higher throughputs than other scenarios. Figure 3: Low Resolution Video Throughput A. Video Throughput Figure 3 and Figure 4 display the video throughput of the four scenarios. DiffServ-MPLS clearly improved the video throughput by more than 40 percent compared to using the conventional routing. DiffServ-MPLS also offers the best video throughput as compared to MPLS and DiffServ only which only gave 20 percent improvement. 965 2009 IEEE Symposium on Industrial Electronics and Applications (ISIEA 2009), October 4-6, 2009, Kuala Lumpur, Malaysia Figure 4: High Resolution Video Throughput B. Video End-to-End Delay As shown in Figure 5 and Figure 6, DiffServ-MPLS had shown the lowest end-to-end delay among these three schemes. In DiffServ-MPLS, the labeling of packets has provided faster processing rate at the routers compared to the conventional IP address matching in routers. It also serves the network in terms of its PHB where prioritization was offered to the video application. Therefore, DiffServ-MPLS demonstrated the advantage to utilize these two attributes in offering the lowest end-to- end delay for video traffic even though at high traffic flow. Figure 5: Low Resolution End-to-End Delay
Figure 6: High Resolution End-to-end Delay C. Other Traffic Throughput Figure 7 and Figure 8 demonstrates how the Assured Forwarded traffic is being affected in the network when using the three QoS schemes. It is reminded that in DiffServ network, FTP was given higher priority than Http traffic. Since FTP was classified better QoS than Http, it is observed that DiffServ-MPLS has served FTP far better than other schemes. This is because in DiffServ-MPLS, FTP and Http shared the same network path while video was routed to another and FTP was given better services than Http. On the other hand, the use of MPLS seems to have benefitted Http traffic more in terms of its throughput. In addition to that, the average throughput of both FTP and Http traffic can be seen to decrease as volume of traffic increases but on the other hand video traffic rate continued to increase. This is because FTP and Http traffic are TCP while video is UDP. In general, TCP source undergoes congestion control phase when congestion is detected in the network and slows down the sending rate while UDP has no congestion control.
Figure 7: FTP Throughput
Figure 8: Http Throughput D. IP Packet Loss Figure 7 and Figure 8 demonstrates how the Assured Forwarded traffic is being affected in the network when using the three QoS schemes. It is reminded that in DiffServ network, FTP was given higher priority than Http traffic. Since FTP was classified better QoS than Http, it is observed that DiffServ-MPLS has served FTP far better than other schemes. TABLE III PACKET DROPS Resolution Packet Drop (packets) Baseline MPLS DiffServ DiffServ MPLS Low 5.7 4.38 5.18 4.79 High 8.56 6.88 8.32 8.35
In DiffServ-MPLS scenario, packet drops were higher than in the MPLS scenario albeit lower than the baseline 966 2009 IEEE Symposium on Industrial Electronics and Applications (ISIEA 2009), October 4-6, 2009, Kuala Lumpur, Malaysia scenario. This is because of the QoS behavior in DiffServ towards FTP and Http traffic where as the buffer overflows, the lowest class will be dropped first. Therefore, increasing the video traffic and giving 75% guaranteed bandwidth to it has its price on other traffic. However, FTP and Http are using TCP protocol which offers error correction that allows lost data to be re-sent until all the packets are successfully transferred. Hence, with this quality and allowing delay in their transmission, some packet drops can be tolerated. Therefore, by permitting a considerate amount of packet drops in DiffServ-MPLS as compared to lower loss in MPLS scenario, DiffServ-MPLS had benefitted the video traffic in the overall performance in terms of delay and throughput. VI. CONCLUSION The combined use of the MPLS DiffServ is envisioned to provide end-to-end guaranteed quality of service (QoS) for multi traffic in IP networks. In this preliminary findings, although, video traffic was given the highest priority in DiffServ and also being routed to other path using MPLS, it still appears to experience some packet drops since video traffic in nature is variable-bit-rate (VBR) and so it will have burst period which may not be able to be handled by the bandwidth available in this topology. However, in this simulation, we have shown that the DiffServ-aware MPLS architecture offers advantages for improved QoS in particular for throughput and delay. Its effect on jitter needs to be investigated for a more comprehensive analysis on the system. Nevertheless there is still a need to study on the QoS mechanism such as traffic policing, queueing, scheduling and congestion avoidance to achieve guaranteed QoS across the IP network. For example, in congestion avoidance, instead of using fixed bandwidth allocation, this scenario can be simulated out using a measurement based bandwidth allocation which will include signaling. Signal is sent from end to end to observe the current bandwidth utilization before allocating the sufficient resources for the traffics. Thus, the bandwidth wastage can be avoided and the accepted traffics will be increased, as the bandwidth allocated which is not utilized by the dedicated traffic can be used by other traffics. On top of that, the work can be extended by simulating point to multipoint that can represent the real network scenario. ACKNOWLEDGMENT Financial support from Ministry of Science, Technology and Innovation (MOSTI) Malaysia ScienceFund No: 02-01-01-SF0006 & 02-01-01-SF0167 is gratefully acknowledged for implementation of this project. Financial assistance of Universiti Teknologi MARA is also gratefully acknowledged. REFERENCES [1] K. Nichols, V. Jacobson, L. Zhang, A Two-bit Differentiated Services Architecture for the Internet, RFC 2638, July 1999 [2] Dongli Zhang, Dan Ionescu: QoS Performance Analysis in Deployment of DiffServ Aware MPLS Traffic Engineering. Eighth ACIS International Conference on Software Engineering, Artificial Intelligence, Networking, and Parallel/Distributed Computing, 2007. SNPD 2007, Page(s): 963-967 [3] Ilias Andrikopoulos and George Pavlou, Supporting Differentiated Services in MPLS Networks, Seventh International Workshop on Quality of Service, IWQoS '99, 1999 [4] E.Rosen et al., Multiprotocol Label Switching Architecture, RFC 3031, January 2001 [5] S.Blake et al.,An architecture for Differentiated Services, RFC2465, December 1998. [6] MPLS Support of Differentiated Services, RFC 3270 [7] J.F.Chiu, Z.P.Huang, C.W.Lo,W.S.Hwang and C.K.Shieh, "Supporting End-to End Qos in DiffServ/MPLS Networks," 10th International Conference on Telecommunications, 2003. ICT 2003. Pages(s): 261-266, 2003 [8] X. Zhu, M. Chen and Y. YU Video Transmission Based on Diffserv Over IP Network, Communication Technology Proceedings, ICCT 2003. International Conference on Vol. 2, pp. 1754- 1757, April 2003 [9] Linawati, I Made Suartika, Self-Similar Traffic Generator, Teknologi Elektro. Vol.4 No. 1 Januari Jun, 2005 [10] Mitko Gospodinov, The effects of different queuing disciplines over FTP, Video and VoIP Performance, Proceedings of the 5th International Conference on Computer systems and Technologies, 2004 [11] Afef Kotti, Rached Hamza, Kamel Bouleimen, New Bandwidth Management Framework for Supporting Differentiated Services in MPLS Networks, International Conference on Communication Software and Networks, 2009. [12] Jasmina Barakovi, Himzo Bajri, Amir Husi, The Impact of Increased Video Traffic on Quality of Service Parameters in Next Generation IP/MPLS Network, 49th International Symposium ELMAR-2007, 12-14 September 2007.