Beruflich Dokumente
Kultur Dokumente
Farrah Farooq
Punjab University College of Information Technology (PUCIT) University of the Punjab Allama Iqbal (Old) Campus, Anarkali, Lahore, Pakistan +92-(0)42-111-923-923
Sidra Aslam
Punjab University College of Information Technology (PUCIT) University of the Punjab Allama Iqbal (Old) Campus, Anarkali, Lahore, Pakistan +92-(0)42-111-923-923
Shahzad Sarwar
Punjab University College of Information Technology (PUCIT) University of the Punjab Allama Iqbal (Old) Campus, Anarkali, Lahore, Pakistan +92-(0)42-111-923-923 Ext-414
farrah.farooq@pucit.edu.pk ABSTRACT
sidra.aslam@pucit.edu.pk
s.sarwar@pucit.edu.pk
Identifying a multicast label-switched-path (LSP) tree that satisfy a set of traffic-oriented and resource-oriented QoS constraints such as cost, reliability, bandwidth, jitter, and delay, has become an important research issue in the area of multicast routing in MPLS networks. In general, multiconstrained multicast tree-selection is an NP-complete problem. In this survey, QoS-based multicast tree-selection algorithms from the perspective of optimization techniques are evaluated. The existing algorithms are classified into three dimensions: non-MPLS-/MPLS-based, single/multiple QoS constrained, and heuristic-/unicast-/artificial-intelligence optimization techniques. In addition to state-of-the-art review of existing solutions, this article highlights important characteristics of QoS-based MPLS multicast algorithms and discuss important issues that are worthy of investigation in future research activities.
General Terms
Algorithms, Performance.
Keywords
Multicasting, MPLS, QoS, Genetic algorithms, Minimum spanning tree, Steiner tree, Artificial intelligence, Ant colony optimization, Feed forward neural network.
IPTV, and its different categories like broadcast TV, video-ondemand (VoD) and near-VoD has created requirements for multicasting through service provider core networks, high availability up-to five-nines (99.999%), QoS, reduced network congestion, and efficient resource utilization. MPLS multicast, at network layer, is the best-fit to these requirements, and major network providers present a strong case for their deployment [1, 2, 3]. There are competing technologies, like IP multicast that has been augmented with protocol extensions [4, 5, 6, 7]. Although it benefits from being widely deployed, it does not provide significant features such as QoS, traffic load distribution, dedicated bandwidth reservation and fast reroutes. In order to satisfy diverse QoS requirements, certain performance objectives must be fulfilled by considering traffic-oriented and resource-oriented constraints. Trafficoriented constraints include packet loss, delay, jitter, and efficient throughput. Whereas, resource-oriented constraints relate to optimization of network assets utilization and its metrics are bandwidth, shortest-distance, and link constraint. Overall, goal is to reduce network congestion, efficient utilization of network resources, minimize network cost, and distribute network load optimally. At network layer, routers generate copies of packets as needed and forward them towards receivers of the multicast group. It requires complex multicast routing protocols and tree-selection algorithm. Multicast tree-selection problem with more than two QoS constraints is considered as optimization problem and is NPcomplete [8]. Recently, a detailed survey on routing optimization for Internet TE is given in [9], but the issue of QoS-based MPLS multicast (QMM) tree-selection algorithms has not been extensively analyzed. In this article, QMM tree-selection algorithms are classified into three dimensional criteria: from the aspect of (1) nonMPLS-/MPLS domain; (2) single/multiple QoS constraints; and (3) heuristic/unicast/artificial-intelligence optimization techniques. An overall classification of multicast tree-selection algorithms is presented in Figure 1, and this article is organized following the structure of this figure. The objective of this article is thus to provide a comprehensive survey with an emphasis on QoS issues, to discuss the current trends and to pave the way for future research direction. The rest of the paper is structured as follows. Section 2 outlines the nonMPLS multicast tree-selection algorithms with single and multiple QoS constraints. The subsequent sections are related
1. INTRODUCTION
The burgeoning Internet multimedia applications, broadcast
Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. FIT09, December 1618, 2009, CIIT, Abbottabad, Pakistan. Copyright 2009 ACM 978-1-60558-642-7/09/12....$10. .
to QMM. Section 3 describes most recent QMM tree selection algorithms. Section 4 discusses the current trends and some of
the future directions for improved QMM. Finally, Section 5 contains concluding remarks.
[12] is the well known algorithm to find MST. In this algorithm, tree construction starts by selecting least cost link connecting an off-tree node to the partial tree. The algorithm is greedy since the tree is augmented with an edge that contributes the minimum amount to the total tree cost. The resultant tree does not guarantee to find the best path from source to each multicast receiver. In Figure 2(c), least cost tree generated by MST algorithm is displayed. Note that the least cost from R1 to R3 using MST is 9, whereas, in SPT algorithm, it was 7.
Figure 2. Multicast tree-selection with single QoS constraint: (a) SPT: if links have unit weights and objective is to find least hop-count tree; (b) SPT: the links have variable weights and objective is to find least cost path for each receiver; (c) MST: the links have variable weights and objective is to find least cost tree. constrained least-cost tree problem. The link cost is defined as a function of link utilization. The link cost is defined as a function of link utilization. BSMA always find a delay constrained multicast tree if one exists because it starts with a least-delay spanning tree. Haberman in [30] considered the Steiner tree problem under the delay and delay jitter constraints. handle traffic-oriented constraints such as end-to-end delay and balance traffic loads in high speed packet switched network. To associate communication networks with the theory of ant algorithm the routing tables are replaced by tables of probabilities called Pherables. They give the probabilities of alternative choices between neighboring links. The idea of algorithm in constructing multicast tree is to find the paths with delay-bounded and less congested degree by launching ant in the source node. The assumption is that each node has the information about spare capacity and propagation delay of its neighboring nodes and links. To establish connection with the entire destinations, algorithm randomly selects a destination node in multicast members set M, and then launch a certain ants with the address of destination. Ant traveling time is recorded and pheromone value is updated locally on the path ant is traveling on. When ants arrive at destination node, if the time ant travels from the source node to the destination node is less than or equal to delay-bound, pheromone value is updated globally on the path. The probability in pherable is re-computed and process is repeated. Using this process one can find the less congested path under bounded-delay. Another technique called Ant-Agent based QoS multicast routing algorithm (AAQMRA) [33] employs the same scheme which has a simpler process and requires less control parameters.
space. That is why the k-shortest path approach is adopted to calculate optimal path. Furthermore, search space is reduced by exploring only non-dominated ones. In the second phase, the sub-graph representing the set of obtained paths is then optimized by eliminating as many cycles as possible without violating the different constraints. Later, in [35] the authors propose taboo QoS multicast routing algorithm (T-QMRA) in which the MAMCRA strategy is used to achieve the first phase . They propose improvements to the second phase of MAMCRA by adopting a meta-heuristic approach based on taboo search algorithm to provide a sub-optimal solution.
LSP; (3) p2m LSPs are established between ingress and multiple egress label switched routers (LSRs). While MPLS provide great flexibility in packet forwarding, a number of issues arise when mapping layer 3 multicast trees onto layer 2 p2m LSPs. [39] discusses these issues in details. Integrating QoS provisioning in MPLS multicast introduce another level of difficulties some of which are discussed in [40]. There has been tremendous work done on MPLS multicast routing protocols [37, 41, 42, 43, 44] to overcome above specified problems but still many of the issues are unresolved. This article does not aim to give MPLS multicast protocols description, although it highlights the underlying p2m LSPs selection algorithms. Following are some characteristics of most of the QMM LSP tree-selection algorithms proposed to date. 1. Distributed algorithm: P2m LSP tree selection and traffic flow assignment requires distributed algorithms that have the ability to handle dynamic changes in the network and can use multiple multicast routing protocols to collect and update network state information. 2. Online algorithm: In the offline/static model, setting up new p2mp LSPs may require reselection of existing p2mp LSPs. This problem may introduce the problem of re-structuring of existing multicast tree, signaling overhead and overconsumed labels. While considering QoS requirements of multiple receivers, solution for this problem is complex. In practice, the multicast algorithm must be an online algorithm capable of selecting p2mp LSP requests in an optimal manner when the requests are not all presented at once, so that future multicast sessions can be handled. This type of online algorithm should handle the interference problem of current LSPs and future p2mp LSP setup requests a priori. 3. Knowledge of ingressegress and intermediate points of LSPs: Even though future demands may be completely unknown, the routers where LSPs can potentially originate and terminate must be known since these are the networks edge routers. Furthermore, algorithm must be able to specify intermediate LSRs and use any available knowledge regarding ingressegress pairs. 4. Aggregated p2m LSPs: These trees come into the category of shared trees (*, G). Aggregation is an approach in which same multicast tree is used for multiple multicast groups. Aggregated trees are built to reduce multicast forwarding states, control messages overhead, and label space. Building such a tree that spans all groups is a difficult issue but it has a significant importance in MPLS. Moreover, QoS based aggregated tree selection is a hot issue in MPLS Multicast. 5. Single QoS constrained: Algorithms consider single QoS resource-oriented constraint such as bandwidth. The authors present strong case for using bandwidth as QoS metric. They think that the most practical way to handle multiple QoS constraints, such as delay, jitter, and losses, is to convert them into efficient bandwidth requirement.
differs from them as it computes multicast tree with single link-constraint. There are various MPLS unicast routing algorithms that can be used in this multicast tree-selection algorithm. The most commonly used algorithm for routing unicast LSPs is the min-hop algorithm [37]. In this algorithm, the path from the ingress to the egress with the least number of feasible links is chosen. Another min-hop-like algorithm widest-shortest-path (WSP) for routing unicast flows which attempts to load-balance the network traffic is proposed in [45].
Figure 3. (a) The ideal multicast tree that fulfills 100 Mb/s bandwidth constraint and avoid future LSP collisions (b) Multicast tree with shortest path algorithm. Link weights are indicating residual bandwidth after LSP utilization which cannot meet the future request (c) Solution: Avoid selecting critical links by increasing their critical weights.
decentralized multicast routing algorithm for nonMPLS/MPLS domains. Another significant possible research direction for MPLS p2m LSP selection problem under multiple QoS constraints is the use of genetic algorithm-based heuristics. Recently in [55], the author presented an analytical study of different selection and crossover methods used in genetic algorithms for the MPLS unicast LSP selection problem. He stated that with exhaustive analysis and properly designed genetic algorithm-based heuristics, MPLS LSP allocation problem can be solved in efficient way. Figure 4 summarizes the evolution of QoS-constrained multicast-tree selection algorithms proposed from early 1980s to the date and indicate future research work in the next upcoming years.
basic working phenomenon with examples. In addition to algorithms analysis, the evolution of multicast tree-selection algorithms from early 1980s is also presented. According to most recent proposed solutions, utilizing ant colony optimization techniques and genetic algorithms for setting-up QoS p2m LSPs in MPLS domain can potentially benefit the internet multimedia applications with high QoS requirements.
6. REFERENCES
[1] S. Natu. Fast Reroute for Triple Play Networks. Tech. Report, Redback Networks, APRICOT, 2006. [2] J. Le Roux. Designing a Carrier Class TV Broadcast Network using P2MP MPLS-TE. Tech. Report, Orange France Telecom Group, 2008. [3] N. Neate. RSVP-TE Point-to-Multipoint: An overview of the RSVP-TE P2MP Extensions, Their Applications to IPTV, and Support for Other (G) MPLS Features. Tech. Report, Data Connection ltd, 2009. [4] J. Moy. RFC-1584: Multicast Extensions to OSPF. IETF, 1994. [5] A. Ballardie. RFC-2189 Core Based Trees (CBT version 2) Multicast Routing. IETF, 1997. [6] T. Bates, and R. Chandra. RFC-2283 Multi-protocol Extensions for BGP-4. IETF, 1998. [7] D. Thaler. RFC-3913 Border Gateway Multicast Protocol (BGMP). IETF, 2004. [8] R.M. Karp. Reducibility among Combinatorial Problems. Complexity of Computer Computations, 85-104, In Plenum Press, 1972. [9] B. Wang and J. C. Hou. Multicast Routing and Its QoS Extension Problems, Algorithms, and Protocols. IEEE Network, 14(1): 22-36, 2000. [10] C. Guoliang, W. Xufa, and Z. Zhanquan,et al. Genetic Algorithms and its Application. Peoples Soft and Telecomm. Press, 1996. [11] S. Pierre and G. Legault. A Genetic Algorithm for Designing Distribute Computer Network Topologies. IEEE Trans. on Manuals, Systems, and Cybernetics, 28(2): 249-258, 1998. [12] T. H. Cormen, C. E. Leirson, and R. L. Rivest. Introduction to Algorithms. MIT Press, 1997. [13] L. Kou, G. Markowsky, and L. Berman. A Fast Algorithm for Steiner Trees. Acta Informatica, 141-45, 1981. [14] F. K. Hwang. Steiner Tree Problems. Networks, 55-89, 1992.
5. CONCLUSION
QoS-based multicast tree-selection problem can be defined as to select a multicast tree, while considering multiple trafficoriented and resource-oriented constraints. It is a NP-hard problem and various multicast tree-selection algorithms have been proposed since 1980 to support group communication over the non-MPLS/MPLS networks with the aim of reducing network resource consumption. These algorithms can be classified into three implementation approaches: unicast routing, heuristic, and artificial intelligence. Most of the traditional algorithms consider only single resource-oriented constraint and static information, to reduce the algorithm complexity. Devising such algorithms that can (1) handle dynamic network topology and membership changes, (2) perform load balancing, and (3) aggregation of multiple group trees, is the topic worthy of investigation within nonMPLS/MPLS domains. Moreover, deterministic computation of QoS-based multicast tree-construction such as shortest-path and unicast-routing-based techniques are usually timeconsuming. For this reason, methods based on artificial intelligence concepts are proposed to solve the complex NP-
Figure 4. Evolution of QoS-based multicast tree-selection algorithms and future research direction. hard problem. Among all these solutions, genetic algorithms and ant colony optimization techniques have emerged as a prevailing technology for solving complex optimization problems of network topologies. This article presents the stateof-the art review of all these algorithms and discusses their
Table 1. Comparison of QoS-based multicast tree-selection algorithms Multicast Network Algorithm SPT MST ST GA-MT NonMPSL CST AA-MT MAMCRA MQMMA MHMA HCCMA MMF L-MMIRA Approach Unicast Heuristic Heuristic Artificial Intelligence Heuristic Artificial Intelligence Unicast Unicast Unicast Heuristic Heuristic Unicast No. of QoS constraints Single Single Dual Multiple Multiple Single Multiple Multiple Single Single Single Single QoS Constraints Link cost (bandwidth or delay) Tree cost Link cost and tree cost Bandwidth, delay, end-to-end delay, jitter Bandwidth, delay, end-to-end delay, jitter Bandwidth or delay Bandwidth, delay, end-to-end delay, jitter Bandwidth, delay, end-to-end delay, jitter Residual link capacity Residual link capacity Residual link capacity Residual link capacity Running Time Complexity O( |L|.log|N| ) O( |L|.log|N| ) O( |M|.|N|2 ) O ( |M|.(|N|+|L|). I.V ) O( |N|4 ) O( C.|N|2.A ) O( k|N|.log(k|N|) + k2m|L| ) O( k|N|.log(k|N|) + k2m|L| ) O(|M|.|L|.log|N|) O( |N3|.log|N| ) O( |L|2.log2|U|) O( 2|L|+|M|.|N|2) Tree type Centralized Centralized Centralized Centralized Centralized Distributed Centralized Centralized Distributed Distributed Distributed Distributed
MPLS
|N|: number of nodes in a network |M|: number of nodes in multicast group |L|: number of links in a network U: maximum link capacity C: number of ant-algorithm cycles
A: number of total ant agents m: number of link constraints k: number of shortest path in each nodal queue I: number of iterations in genetic algorithm V: number of chromosomes in each population Management of Multimedia Networks and Services, 3271: 140-151, 2004.
[15] B. Wang and J. C. Hou. Multicast Routing and Its QoS Extension Problems, Algorithms, and Protocols. IEEE Network, 14(1): 22-36, 2000. [16] H. Takahashi and A. Matsuyama. An Approximate Algorithm for Steiner Tree Problem in Graphs. Math. Japanica, 24: 573-77, 1980. [17] N. F. Maxemchuk. Video Distribution on Multicast Networks. IEEE JSAC, 357-84, 1997. [18] J. Hesser, R. Manner, and O. Stucky. Optimization of Steiner trees Using Genetic Algorithms. In Proc. of Int. Conf. on Genetic Algorithms, 231-236, 1989. [19] B. A. Julstrom. A Genetic Algorithm for Rectilinear Steiner Problem. In Proc. of Int. Conf. on Genetic Algorithms, 474-480, 1993. [20] Y. Leung and G. L. Zong-Ben. A genetic Algorithm for the Multiple Destination Routing Problems. IEEE Trans. on Evolutionary Computation, 2(4): 150-161, 1998. [21] F. Xiang, L. Junzhou, W. Jieyi, and G. Guanqun. QoS Routing Based on Genetic Algorithm. Computer Communications, 22(15), 1392-1399, 1999. [22] W. Zhengying, S. Bingxin, and Z. Erdun. BandwidthDelay-Constrained Least-Cost Multicast Routing Based on Heuristic Genetic Algorithm. Computer Communications, 24(7-8): 685-692, 2001. [23] N. Wang and G. Pavlou. Bandwidth Constrained IP Multicast Traffic Engineering without MPLS Overlay.
[24] V. P. Kompella, J. C. Pasquale, and G. C. Polyzos. Multicast Routing for Multimedia Communications. IEEE TON, 1(3): 286-92, 1993. [25] Q. Zhu, M. Parsa, and J. Garcia. A Source-Based Algorithm for Delay-Constrained Minimum-Cost Multicasting. IEEE INFOCOM, 1: 377-84, 1995. [26] B. K. Haberman, and G. Rouskas. Cost Delay, and Delay Variation Conscious Multicast Routing. Tech. Report, TR-97-03, 1997. [27] M. Dorigo and L. M. Gambardella. Ant Colony System: A Cooperative Learning Approach to the Traveling Salesman Problem. IEEE Trans. on Evolutionary Computation, 1(1): 53-65, 1997. [28] L. Guoying, L. Zemin, and Z. Zheng. Multicast Routing Based on Ant Algorithm for Delay-Bounded and LoadBalancing Traffic. IEEE LCN, 2000. [29] X. Yan and L. Li. Ant Agent-Based QoS Multicast Routing in Networks with Imprecise State Information. In Proc. of Inter. Conf. PRIMA, 374-385, 2006. [30] P. Kuipers and M. Van. Constrained-Based Multicast Routing Algorithm. Computer Communications, 25 (8): 801-810, 2002. [31] N. Ben Ali, A. Belghith, J. Moulierac, and M. Molnar. QoS Multicast Aggregation under Multiple Additive Constraints. Computer Communications 31(15): 3564 3578, 2008.
[32] N. Ben Ali, M. Molnar, and A. Belghith. MultiConstrained QoS Multicast Routing Optimization. INRIA Research Report, 2008. [33] D. Ooms and W. Livens. IP Multicast in MPLS Networks. In Proceedings of International Conference, High Performance Switching and Routing, 301-305, 2000. [34] O. Awduche, L. Berger, D. Gan, T. Li, V. Srinivasan, and G. Swallow. RFC-3209 Extensions to RSVP for LSP Tunnels. IETF, 2001. [35] Y. Zhou and Y. Zhu. Active Ideal Multicast Tree in MPLS Environment. In Proc. of Int. Conf., ALPIT, 583588, 2007. [36] B. Yang and P. Mohapatra. Multicasting in MPLS Domain. Computer Communications, 27(2): 162-70, 2004. [37] D. Ooms, B. Sales, W. Livens, A. Acharya, F. Griffoul, and F. Ansari. RFC-3353 Overview of IP Multicast in MPLS Environment. IETF, 2002. [38] R. Aggarwal, D. Papadimitriou, and S. Yasukawa. RFC4875 Extensions to Resource Reservation Protocol-Traffic Engineering (RSVP-TE) for Point-to-Multipoint TE Label Switched Paths (LSPs). IETF, 2007. [39] T. Eckert, E. Rosen, R. Aggarwal and Y. Rekhter. RFC 5332 - MPLS Multicast Encapsulations. IETF, 2008. [40] M. Kodialam and T. V. Lakshman. Minimum Interference Routing with Applications to MPLS Traffic Engineering. INFOCOM, 2000. [41] R. Guerin, D. Williams, and A. Orda. QoS routing mechanisms and OSPF extensions. GLOBECOM, 1997. [42] Y. Seok, Y. Lee, Y. Choi, and C. Kim. Explicit Multicast Routing Algorithm for Constrained Traffic Engineering. IEEE ISCC, 2002. [43] M. Kodialam, T. V. Lakshman, and S. Sengupta. Online Multicast Routing With Bandwidth Guarantees: A New Approach Using Multicast Network Flow. IEEE TON, 11(4): 676-686, 2003. [44] M. Charikar, C. Chekuri, T. Cheung, Z. Dai, A. Goel, S. Guha, and M. Li. Approximation Algorithms for Directed Steiner Problems. In Proc. of ACM/SIAM Symp., Discrete Algorithms, 192-200, 1998. [45] R. K. Ahuja, T. L. Magnanti, and J.B. Orlin. Network Flows: Theory, Algorithms, and Applications. Englewood Cliffs, NJ: Prentice-Hall, 1993. [46] L. Xuan and L. Ying. L-MMIRA: Light Multicast Minimal Interference Routing Module in MPLS Network. IEEE ICN, 421-426, 2008. [47] M. Dorigo, V. Maniezzo and A. Colorni. The ant system: Optimization by a Colony of Cooperating Agents. IEEE Trans. on Systems, Man, and Cybernetics-Part B, 26(1): 29-41, 1996. [48] A. A. Turky and A. M. Thiel. Prediction-Based Decentralized Routing Algorithm. Electronic Communications, EASST, 17, 2009.
[49] E. Einhorn and A. Mitschele-Thiel. RLTE: Reinforcement Learning for Traffic-Engineering. In Proc. of Int. Conf., AIMS, 2008. [50] A. Eswaradass, X. H. Sun, and M. Wu. Network Bandwidth Predictor (NBP): A System for Online Network performance Forecasting, In IEEE Int. Sym., Cluster Computing and the Grid, 265-268, 2006. [51] A. V. Andrade, L. Errico, A. L. L. Aquino, L. P. Assis, and C. H. N. R. Barbosa. Analysis of Selection and Crossover Methods used by genetic Algorithm-based Heuristic to solve the LSP Allocation Problem in MPLS Networks under Capacity Constraints. In Proc. of Int. Conf., EngOpt, 2008.