Multicast Routing in ATM Networks with Multiple Classes of QoS Ren-Hung Hwang, Min-Xiou Chen, and Youn-Chen Sun Department of Computer Science & Information Engineering National Chung Cheng University
Outline Introduction to multiple QoS constraints multicast routing problem Network Model QoS Decomposition Algorithm Heuristic Multicast Routing Algorithm Conclusion
Introduction Multimedia applications provide multiparty communication services recently Multimedia data need to be delivered with end-to-end performance guarantee Data forwarding on Internet is based on best effort without QoS guarantee Providing QoS requires cooperation of many control mechanisms
Single QoS constrained problem The problem of routing multicast traffic with one QoS constraint Heuristics compute low-cost multicast trees which guarantee an upper bound delay Each link is assumed to be associated with a set of QoS metric ex : (2, 2, 1) 2 means delay of the link 2 means required bandwidth of the link 1 means cost of the link
Multiple QoS classes routing problem Current ATM switches or Internet routers support multiple classes of QoS End-to-end QoS number provided by a path is equal to K |P| K : number of QoS classes on each link |P|: length of the path Routing algorithm not only need to find a constrained minimum cost tree,but also need to find optimal local QoS allocation
Solution to this problem Assume each links associated with k QoS classes,and each class has four QoS metrics Two tasks: QoS decomposition use greedy algorithm to decompose end-to-end QoS constraint QoS-constrained multicast routing extending three single-constraint Steiner tree algorithms ‧ CSPT ‧ MST ‧ BSMA
Network Model An ATM network is represented by G(V,E) Link from node i to j is denoted by e(i,j) Each link e has K set of QoS metrics The QoS guarantee provided by kth set of QoS metrics on link e is denoted by: Q e ( d k,l k,b k,c k ) Bandwidth required for certain level of QoS can be computed by effective BW technique Cost function is based on COL
Network Model (cont.) A multicast connection request consists of three parameters : ( s, D, Q ) s : source node D : set of destination nodes Q : set of end-to-end QoS requirements {s} ∪ D be a multicast group and let ω be the multicast group Each connection is associated with a reward parameter representing the revenue received for carrying this connection
Mathematical description A path P with links e 1,e 2,……,e p Let local QoS classes selected on these links are k 1,k 2,……,k p Q e (d k ), Q e (l k ), Q e (b k ), Q e (c k ) denotes the value of delay, loss prob., bandwidth and cost Then the end-to-end QoS can be satisfied if : Q e i ( d k i ) ≦ D 1 ﹣ (1- Q e i ( l k i )) ≦ L
Performance Metric We adopt fractional reward loss as the performance metric FRL is defined as : FRL = Minimizing the fractional reward loss is equivalent to maximizing the expected revenues produced by the network
QoS Decomposition Algorithm The idea of greedy algorithm The detail description 1. Feasibility check : a. when all QoS classes are set to 0 b. has sufficient bandwidth to provide loosest QoS class 2. Initialize : set local QoS class on each link to the loosest QoS level i.e : k e i = K, 1 ≦ i ≦ p
QoS Decomposition Algorithm(cont.) 3. Loop a. If end-to-end QoS requirement is met,return success b. no link has sufficient bandwidth,return fail c. Case 1: if both end-to-end delay and loss probability is not met by current QoS allocation Case 2: depends on which QoS metric cannot be met d. select the link that has minimum cost for adjusting up QoS level to the desired level 4. End loop
Heuristic Multicast Routing Algorithm Extend three heuristic algorithms proposed for single-constraint multicasting problem to find trees with multiple QoS classes Constrained shortest path tree (CSPT) Algorithm proposed by Kompella (KPP) Bounded shortest multicast algorithm (BSMA) All of them require an algorithm to find a constrained path between two nodes CSP algorithm
CSP algorithm for point-to-point QoS routing CSP algorithm finds a constrained minimum cost path from source to each destination Let.cost[d][l] be the minimum cost from the source to node i when delay is d and loss probability l Detail description : 1. Initial phase :,,for all possible d,l
CSP algorithm 2. Loop For all i V - {s } = ( ) until
CSPT Multicast Algorithm CSPT algorithm first uses CSP algorithm to find a constrained minimum cost path A multicast tree is then form by concatenating these paths Two issues needed to be noticed : These paths may share some common links allocate tightest QoS to these common links Looping remove some of the paths until no cycle is found
Example S AB D1 D2 d=2 d=4 d=2 d=1 d=3 d=1 Common link problem d=2 d=1 d=2 S D1 D2 Looping problem
The MST algorithm In the first step,using CSP algorithm to find all-pair constrained cheapest path After first step,a fully connected closure graph can be formed In the second step,the algorithm repeatedly selects a destination with minimum cost to join the tree case1: destination node is connected to source node case2: destination node is connected to another destination node already on the tree
Original BSMA Algorithm Original BSMA algorithm first constructs a minimum spanning tree T (s,M) It repeatedly replaces “superedges” in T with lower-cost “superedges”,until cost can not be reduced any further Remove superedge results in two subtrees, and BSMA then finds a “delay-bound shortest path” to reconnect two subtrees
Modified BSMA Algorithm Using CSPT algorithm to construct a multicast tree satisfying end-to-end QoS of each O-D pair first In second step,superedges are selected and replaced with lower cost path repeatedly Modified BSMA algorithm uses the CSP algorithm to find a path with lower cost for reconnecting these two subtrees
QoS classes and effective bandwidth 4 QoS class Delay Loss probability Required bandwidth
20-node random graph
Performance of Multicast Algorithms
Conclusion and Future Work Simulation results show that BSMA algorithm yields better performance than other algorithms The proposed multicast algorithms are centralized algorithms Computational complexity of the proposed CSP algorithm is quite high due to the large number of QoS classes