Sheeja S et al,int.j.computer Technology & Applications,Vol 3 (2),

Transcription

1 Performance Analysis of Genetic CR selection in Multicast Multirate Network Sheeja.S # 1, Dr.Ramachandra V. Pujeri*2 # Research scholar, Bharathiar University, Coimbatore * Vice Principal, KGISL Institute of Technology, Saravanampatti, Coimbatore sheejakhader@yahoo.com Abstract This paper presents a performance analysis of different methods used to select the congestion representative over the genetic algorithm based Congestion Representative selection in a multicast multirate network. A huge volume of control traffic is introduced into the network, and the routers are heavily loaded because the entire rate control burden has been shifted to them. Therefore, the number of these operations is limited during a period and restricts the effectiveness of these multicast congestion control methods. In multirate technique, the source maintains several layers each with different transmission rate, and receivers give to different subsets of these layers depending on their and network s bandwidth and congestion circumstances. In a multi-rate multicast session, each layer uses a separate multicast group address. On the whole operations and performance of MCC schemes deal with quite a lot of issues in achieving a wellperforming multicast such as congestion representative (CR) selection. CR Selection based on the genetic algorithm is made so that a optimized representative is selected and fully dynamic order of joining the layers is achieved and performance is much better than the other conventional methods. Keywords congestion control, multirate, multicast, congestion representative, genetic algorithm 1. Introduction Multi-hop communication networks are expected to efficiently serve many applications with diverse characteristics and quality-of-service constraints. Many of these applications, including voice/video broadcast and file sharing, generate data to be multicast to a group of destinations rather than a single one. Traditionally, the rate of such a single-rate multicast session is chosen such that all the receivers can successfully receive the transmitted information at the selected single rate. In a multicast environment, this solution tends to be suboptimal because there is no single target rate for a group of heterogeneous receivers. The problem with using single-rate multicasting strategy is that its throughput is limited by the capacity to the bottleneck destination. In order to overcome this problem, we study multi-rate multicast capabilities, where the source is allowed to multicast its data to different destinations at different rates based on the condition of the network to them. With this capability, the bottleneck destinations will not able to throttle the whole communication quality. Thus, multirate multicast transmission meets the diverse bandwidth requirements of a multicast group by enabling higher data rates to receivers with higher link capacities. Given the multirate multicast nature of transmission, we further require a dynamic scheme that finds optimal scheduling and routing solution for sending the information through the network. The motivation for such a dynamic algorithm stems from the following reasons. Topologies of communication networks constantly vary due to periodic failure of links and routers. Compositions of multicast groups also change frequently due to joining and leaving of receivers. In real time traffic scenarios, the bandwidth available for serving real time traffic also vary depending on the amount of data traffic which change rapidly and may not be readily known at the link scheduler. Thus we need a practical multirate multicasting algorithm which not only finds the right amount of data to be injected in to the network but also dynamically finds optimal routes, scheduling decisions and coding solution to route the multicast traffic. All earlier multi-layer schemes such as GMCC [9] allows only a predefined order of joining the layers, i.e. a receiver can join layers 1, 2 and 3 in sequence, but cannot join layers 1 and 3 without joining the layer 2 in between. Unsatisfied receivers join a new layer if they detect that they are significantly less congested than the CR for their highest layers. Thus goal of this work is to develop CR Selection based on the genetic algorithm is made so that a optimized representative is selected and fully dynamic order of joining the layers is achieved and performance is much better than the other conventional methods. The paper is organized as follows. The problem description is formally in Section 2, Literature survey is kept in section 2, Overview of the proposed work is in 626

2 Section 4, Performance Analysis is kept in Section 5 and the paper concludes with a summary of the results in Section Problem description In the existing multicast congestion control system the receiver of the worst congestion level is selected as the congestion representative, based on which all decisions to join or leave a multicast group so called layers is made. And the transmission rate of the sender is adjusted to TCP throughput of the representative. Though this method is tcp friendly this technique leads to packet loss because of the rapid change of the representative on some mere assumption. In particular, TFMCC [3] rate adaptation and congestion representative (CR) selection are totally left to the single-rate MCC scheme that is being used. Similarly, creation and management of feedback packets at the receivers are done by the single-rate MCC scheme. All earlier multi-layer schemes such as GMCC [9] allows only a predefined order of joining the layers, i.e. a receiver can join layers 1, 2 and 3 in sequence, but cannot join layers 1 and 3 without joining the layer 2 in between. Unsatisfied receivers join a new layer if they detect that they are significantly less congested than the CR for their highest layers. 3. LITERATURE SURVEY A. Sinlge Rate Schemes The RLM [4] and PLM [5] or dynamic but are defined by a carefully designed pattern, such as in RLC [6], FLID-DL recipients have to increase or decrease their receiving rates by joining or leaving some groups, to perform join and leave operations, they send IGMP messages to routers. Upon the receipt of these IGMP messages, routers update their multicast group states to allow traffic through (for join) or stop traffic forwarding (for leave), which allows adjusting throughput for receivers. To quickly react to congestion, these operations have to occur frequently. As a result, a large volume of control traffic is introduced into the network, and the routers are heavily loaded because the entire rate control burden has been shifted to them. Therefore, the number of these operations is limited during a period and restricts the effectiveness of these multicast congestion control schemes. B. TFMCC In TFMCC [3], the receiver that the sender believes currently has the lowest expected throughput of the group is selected as the current limiting receiver (CLR). The CLR sends continuous, immediate feedback to the sender without any suppression, so the sender can use the CLR s feedback to adjust the transmission rate. In addition, any receiver whose expected throughput is lower than the sender s current rate sends a feedback message, and to avoid feedback implosion, biased feedback timers in favour of receivers with lower rates are used C. Smooth Multicast Congestion Control SMCC [7] is a hybrid of single-rate and multi- rate multicast congestion control. It combines a single-rate scheme TFMCC with the receiver-driven idea. In each layer, the source adjusts sending rate within a certain limit based on TFMCC, and receivers join or leave layers cumulatively according to their estimated maximum receiving rates using TCP throughput formula. Since the flows in each layer are adaptive to network status, the number of join and leave operations are greatly reduced. The congestion control is more effective. SMCC [7] requires static configuration of the maximum sending rates for each layer. This requirement makes SMCC not capable of accommodating receivers with variant bandwidth circumstances. In the case when many or all of the receivers fall into the lowest layer, SMCC cannot provide new layers with smaller sending rates. Again, when many or all of the receivers subscribe to the very highest layer(s), then lower layers become redundant, thereby causing the scheme to spend extra effort to maintain those unnecessary layers. D. GMCC The functions of the source and the receivers in GMCC[9] can be decoupled into two main categories: intra-layer, and inter-layer. GMCC uses single-rate MCC to manage intra-layer activities at the source and the receivers. In particular, rate adaptation and congestion representative (CR) selection are totally left to the single-rate MCC scheme that is being used. Similarly, creation and management of feedback packets at the receivers are done by the single-rate MCC scheme. GMCC performs layer join and leave operations at receivers by using statistical measures such as throughput attenuation factor (TAF). Similar to all earlier multi-layer schemes, GMCC allows only a predefined order of joining the layers, i.e. a receiver can join layers 1, 2 and 3 in sequence, but cannot join layers 1 and 3 without joining the layer 2 in between. Unsatisfied receivers join a new layer if they detect that they are significantly less congested than the CR for their highest layers. In particular, for a receiver i having 627

3 a highest layer j, the receiver i joins a new layer if its TAF is significantly smaller than the TAF of the CR for layer j. GMCC does not allow join attempts and join decisions are made purely by the receiver thereby simplifying the operations significantly. Once the receiver joins a new layer it will start participating in the CR selection process for its new highest layer and maybe will be selected as the new CR. When a receiver in GMCC is selected as the CR of a layer, it checks whether or not it is the CR of its highest two layers. If so, then that receiver unsubscribes from its highest layer. In order to dynamically adjust the number of layers GMCC performs layer control by activating or shutting down layers without setting a particular sending rate range for individual layers. In order to implement the layer control, GMCC leverages the fact that CR of each individual layer sends feedback packets regularly to help the source adapt the layer sending rates. In layer control, two operations can happen: (i) activation of a previously empty layer, and (ii) deactivation of a layer. The activation operation happens only when a receiver joins a layer, which did not have any receiver before. From regular CR statistics conveyed by the source, the newly joining receiver realizes that there is no CR for this layer and starts sending feedback thinking it is the CR of the layer. The source, then, figures out that there is a new receiver for the layer and activates the layer. The deactivation operation takes place when the last receiver leaves the layer. Since it is the last receiver in the layer, it must be the CR of the layer. CR of each layer regularly sends feedback to the source for rate adaptation. Once the last receiver and the CR leave a layer, the source will not receive these feedback packets. It will time out and ask receivers in the layer to elect a new CR, which will not occur since no other receiver is left in the layer. In that case the sources will time out for the whole layer and stop sending the data packets thereby shutting down the layer. While earlier schemes like SMCC [7] fix the sending rate ranges of layers as well as the number of layers, GMCC provides the flexibility of varying them. This characteristic of GMCC is very useful for data streaming applications over highly heterogeneous set of receivers, e.g. multicasting multimedia content to very large number of users located at different parts of the Internet. The representative selection mechanism in GMCC scheme is simplistic, but there is certain complexity involved in generating CC. The representative set is not guaranteed to include the slowest receiver, which means that the slowest receiver can be overloaded. Furthermore, it assumes that only a few bottlenecks cause most of the congestion. Based on this assumption, receiver suppression is the only mechanism for filtering feedback from receivers. In a heterogeneous network, where there may be many different bottlenecks and asynchronous congestion, the assumption may not be true. Consequently, the transmission rate may be reduced more than necessarily and stay very low or close to zero. This is known as the drop-to-zero problem. Although PGMCC [2], TFMCC and MDP-CC use different policies for rate adaptation, they all leverage the TCP throughput formula for allocating the slowest receiver, i.e. the receiver with the lowest estimate TCP throughput according to the formula. Therefore, it is necessary for them to measure packet loss rate and RTT for all receivers. The PGMCC sources measures RTT between itself and all receivers in terms of packet numbers, and compare the estimated throughput for updating acker. Due to the necessity of RTT measurement for all receivers, feedback suppression may have serious effect on PGMCC performance. In fact, PGMCC does not provide a feedback suppression mechanism. TFMCC [3] adjusts the rate according to the estimated rate calculated by the representative. The sender needs to echo receiver s feedback according to some priority order, and there is one-way delay RTT adjustment plus sender-side RTT measurement. TFMCC comes with feedback suppression that is an enhanced version and is probabilistic timer-based. Therefore, the total number of feedbacks is the function of the estimated total number of receivers, and additional delay is introduced into feedback. Similar to TFMCC, the target rate is also calculated by the representative. In contrast to PGMCC and TFMCC, MDP-CC maintains a pool of representative candidates for representative update, maintaining multiple representative candidates requires much effort. MDP- CC can use probabilistic timer-based feedback suppression that has the same properties as that of TFMCC. The capability of adjustment at both sides (sender and receiver) is again a complicated process. In each layer, the source chooses a most congested receiver as congestion representative (CR) and adjusts the sending rate of this layer according to the CRs feedback. The source starts traffic in a layer when the first receiver joins and stops traffic in a layer when the last receiver leaves. Each receiver joins layers cumulatively, and is allowed to be the CR of at most one layer. When a receiver detects that it is much less congested than the most congested receiver (i.e. the CR) in the highest layer it has joined, meaning it can potentially receive at a higher rate, it joins an additional layer successively When a receiver detects that it is the most congested receiver in more than one layer, which means it confines or can potentially confine the sending 628

4 rates of more than one layer, it leaves the highest joined layer. Receivers make decisions of join and leave For CR Selection, the GMCC source participates in the messaging and maintains the actual list of CRs pertaining to each layer. Depending on the single- rate MCC scheme being used, the CR selection mechanism can differ. However, in order for interlayer operations to work, GMCC [9] requires the source to maintain the list of current CRs. The Source piggybacks the data packets for various control information to be conveyed to the receivers. Data Packet Handler s main job is to piggy-back this intra-layer information on the multicast session s data packets. Per-layer information that needs to be stored at the source is the list of necessary layer statistics needed to decide about layer control operations. When the CR detects packet loss, it sends feedback packets called congestion indications (CIs) back to the source that decreases the sending rate by half. Data packet GA based CR Selection Optimized CR Dynamic Layer Feed back 4. Proposed work overview The existence of a congestion representative (CR) with explicit feedbacks corresponding to congestion indications (CIs). Note that a CR with CIs is a basic component of all existing single-rate MCC schemes. CR Selection based on the genetic algorithm is made so that a optimized representative is selected and fully dynamic order of joining the layers, i.e. a receiver can join layers any layer which is suitable not in sequence. This method enters routes between nodes only when the routes are requested by the source nodes giving the network the freedom to allow nodes to enter and leave the network at will. Routers remain active only as long as data packets are travelling along the paths from the source to the destination. When the source stops sending packets the path will time out and close. A path for each destination is selected and coded into a single chromosome fitness values are computed with the addition of hops from the source to each destination, similarly the addition of delays from the source to each destination is done. For each link used, the addition of the bandwidth the flow requires for transmission or the total link bandwidth in case the flow exceeds it. Link maximum utilization along the entire path. Put chromosomes in the non-dominated set or discard it depending of their fitness values. Do the cross over. For this, entire paths from different chromosomes are selected and swapped. Note that the paths are not changed but the multicast distribution tree does. Do the mutation and changed entirely by searching a new path in the network modelled by the optimal CR found. Fig.1 Block Diagram of optimized CR selection packet 5. Performance Analysis A. Simulation Environment The simulation experiment is carried out in LINUX (FEDORA 6). The detailed simulation model is based on network simulator-2[17] is used in the evaluation. The NS instructions can be used to define the topology structure of the network and the motion mode of the nodes, to configure the service source and the receiver, to create the statistical data track file and so on. We make all sources ECN-capable, and set the maximum window size to be 1000 instead of 64, so the maximum window size will not be a threshold for the flow rate in all scenarios. For most simulations in this section, we consider a network with a single bottleneck link and multiple flows, as shown in Fig. 2. Since there is a large population of users in the Internet and the fluid model makes sense only when the number of users is large, we choose a sufficiently large number of users for most of the simulations. 629

5 No.of Packets delivered Sheeja S et al,int.j.computer Technology & Applications,Vol 3 (2), S 2 S 1 D 1 D 2 suitable not in sequence load balancing is highly ensured as shown in Fig. 4. S 3 d R 1 R 2 D 3 S N D N Fig. 2 The network topology B. End-End Average Packet delivery ratio The average end-to-end packet delivery was improved in proposed method as compared to both GMCC and SMCC. Fig. 3 shows the better performance. As described in the earlier section, GMCC transmission rate of the sender is adjusted to TCP throughput of the representative. Though this method is tcp friendly this technique leads to packet loss because of the rapid change of the representative on some mere assumption and mmaintains only one route per destination and consequently, each packet in that layer is unable to deliver is dropped since there are no alternate routes SMCC GMCC Proposed Algo Fig. 4 Load balance for 30 sources D. Throughput In figure 5 graphs plotted between throughputs and time which shows improvement in throughput of proposed approach with respect to existing schemes. We can observe from graph that proposed approach is better than existing ones because we are taking optimal decision according to adaptive throughput which helps to increase the throughput due to joining multilayer instead of single layer Here at time 40ms, our approach throughput is 87% more than SMCC which is 50% and GMCC gets 70%. Time(ms) Fig. 3 Ratio of End-End packet delivered C. Routing Load Balance Comparison The proposed algorithm out performs when compared with SMCC and GMCC. When the number of sources is low, the performance of SMCC and Proposed work is similar regardless of mobility. With large numbers of sources, proposed algorithm starts outperforming for higher network traffic. Since optimized representative is selected and fully dynamic order of joining the layers, i.e. a receiver can join layers any layer which is Fig. 5 Throughput comparison of SMCC, GMCC and Proposed Algo 6. Conclusions The proposed work is a new and efficient scheme of multirate multicast congestion control. Theoretical analysis of the proposed scheme is dynamic and good. 630

6 The bandwidth is fairly shared among the heterogeneous users. In general from the simulation it is observed that our proposed scheme achieves high packet delivery ratio. Low bandwidth usage, minimum response time over other conventional schemes. 7. References [1] S.Floyd and V. Jacobson, Random early detection for congestion avoidance, IEEE/ACM Transactions Networking, vol. 1, pp , July [2] L. Rizzo, PGMCC: A TCP-friendly single-rate multicast congestion control scheme, in: Proceedings of ACM SIGCOMM, [3] J. Widmer, M. Handley, Extending equation-based congestion control to multicast applications, in: Proceedings of ACM SIGCOMM, [4] S. McCanne, V. Jacobson, M. Vetterli, Receiver-driven layered multicast, in: Proceedings of ACMSIGCOMM, [5] A. Legout, E. Biersack, PLM: fast convergence for cumulative layered multicast transmission schemes, in: Proceedings of ACM SIGMETRICS, [6] L. Vicisano, L. Rizzo, J. Crowcroft, TCP-like congestion control for layered multicast data transfer, in: Proceedings of IEEE INFOCOM, April [7] G.-I. Kwon, J. Byers, Smooth multirate multicast congestion control, in: Proceedings of IEEE INFOCOM, April [8] Stevens, W., TCP Slow Start, Congestion Avoidance, Fast Retransmit, and Fast Recovery Algorithms, RFC 2001, January [9] Jiang Li, Murat Yuksel, Xingzhe Fan and Shivkumar Kalyanaraman, Generalized Multicast Congestion Control, Computer Networks, Elsevier Science, Volume 51, Pages , August [10] W. Feng. D.Kadlur, D. Saha, and K. Shin, The Blue active queue management, IEEE/ACM transactions on networking, vol 10, No.4, August [11] V.Jacobson,Congestion avoidance and control, in Proc. ACM SIGCOMM, Aug 1998, pp [12] S.Floyd, R.Gummadi, and S.Shenker, Adaptive RED: An algorithm for increasing the robustness of RED s active queue management, Technical Report, Aug. 1,2001. [13] K. Kar, S. Sarkar, L. Tassiulas, Optimization based rate control for multirate multicast sessions, in: Proceedings of IEEE INFOCOM, April [14] L.S. Brakmo, S.W. O Malley, and L.L. Peterson. TCP Vegas: New techniques for congestion detection and avoidance, ACM SIGCOMM Conference, pages 24 35, May [15] Zhang QF, Leung YW. An orthogonal genetic algorithm for multimedia multicast routing. IEEE Transactions on Evolutionary Computation 1999;3: [16] K. Ramakrishnan and S. Floyd, A proposal to add Explicit Congestion Notification (ECN) to IP, RFC 2481, Jan [17] NS-2, The ns Manual (formally known as NS Documentation) available at http: //www. isi.edu/nsnam/ns/doc. 8. Author name(s) and affiliation(s) S.Sheeja Research scholar, Bharathiar University, Coimbatore. She received the MCA degree from Bharathiar University, Coimbatore in She received M.Phil degree in Computer Science from Bharathiar University, Coimbatore in She has 8 years of teaching experience. She is currently pursuing doctoral research in computer networks. At present she is working as Assist Professor in Computer Applications Department at Karpagam University, Coimbatore. Dr.Ramachandra V. Pujeri Vice Principal, KGISL Institute of Technology, Saravanampatti, Coimbatore. The Placement Centre is headed by him. He possesses valuable experience in computer science and other functions in Industry Majors. 631