1 A Distributed Delay-Constrained Dynamic Multicast Routing Algorithm Quan Sun and Horst Langendorfer Telecommunication Systems Journal, vol.11, p.47~58, 1999
2 Outline Motivation Network model and problem definition DCR algorithm DCDMR algorithm Performance evaluation Conclusion
3 Motivation Problem –Multicast group members always change frequently –Few delay-constrained multicast routing algorithm support dynamic groups –Most existing dynamic routing heuristics only consider one link metric
4 Motivation (cont) Objective –Find a dynamic multicast routing algorithm with characteristics : Accommodate members change Satisfy end-to-end QoS requirement Efficient network resource utilization Scale well
5 Outline Motivation Network model and problem definition DCR algorithm DCDMR algorithm Performance evaluation Conclusion
6 Network Model Given –Connected, directed graph G(V,E) –Associated with each asymmetrical link Delay D(e) Cost C(e) –Source node s ( ) –Set of destination S ( ) –Delay tolerance
7 Problem Find –A tree T ( ) rooted at s and span all nodes in S satisfying – is minimum –
8 Model of Dynamic Routing Problem Given –Directed network G(V,E) –Non-negative link metric with Link costs C and link delays D –Delay tolerance –Initial delay-constrained tree with source s ( ) –Series requests
9 Dynamic Problem Find –A series of delay-constrained multicast tree Members of T i are T 0 modified by requests Cost of T i is minimum of all possible T i
10 Outline Motivation Network model and problem definition DCR algorithm DCDMR algorithm Performance evaluation Conclusion
11 Routing Information Each node –Know delay of all outgoing links –Maintain a delay and cost vector with |V|-1 entries Assume delay and cost vectors do not change during execution routing algorithm
12 Delay and Cost Vector Entry for in delay vector of –Destination ID : –End to end delay : –Cost of least delay path : –Next hop of on : Entry for in cost vector of –Destination ID : –End to end delay : –Cost of least cost path : –Next hop of on :
13 Delay-Constrained Unicast Routing Algorithm (DCR) Objective –Construct a low-cost path from s to d satisfying given delay bound Idea –First check if –Always try to along least cost path toward d if –until d is reached
14 DCR Algorithm If exit True If False Next_node= path_direction=LC path_construction=(d,LC) True False Next_node= path_direction=LD delay_so_far= path_construction=
15 DCR Algorithm (cont) If Successful construct path False True If Path_direct=LC or delay_so_far+ False Next_node= path_direction=LD delay_so_far= delay_so_far + path_construction= Next_node= path_direction=LC path_construction=(d,LC) True
16 Example of DCR algorithm A C B D E (30,1) (20,2) (10,1) (10,2) (30,1) Source node : A Destination node : E Delay bound : 5 (cost,delay) pair with each link A C E D B (30,1) (10,2) (10,1) (20,2) A C E D B (10,1) (30,1) (10,2) (10,1) A C D E B (20,2) (30,1) (10,2) (10,1)
17 Outline Motivation Network model and problem definition DCR algorithm DCDMR algorithm Performance evaluation Conclusion
18 Delay-Constrained Dynamic Multicast Routing (DCDMR) DCDMR –Based on DCR algorithm – denote delay of unique path from s to on existing tree T Idea of FAST mode –New member contacts on tree node with request (,mode) along least delay path
19 DCDMR with FAST Mode and If False exit If True False Send Req to v p ’s parent Use DCR to compute a path from v p to v n satisfying constraint True
20 DCDMR with SLOW Mode If and False exit True If False Send Req to v p ’s parent True If True Rsp=(v n,SLOW,v p, ) and send to all v p ’s neighbors False Rsp=(v n,SLOW,v p, ) and send to all v p ’s neighbors
21 DCDMR with SLOW Mode (cont) If False Send Rsp to v n along least delay path True If True False Rsp is updated and sent to all neighbors except sending node True
22 SLOW mode (cont) New member will receive set of responses –Selects one with least cost value –Assume selected one is (,SLOW,, ) –Sends request to – uses DCR to compute a path to satisfying delay constraint
23 Example of DCDMR with SLOW Mode s VpVp VnVn Req Existing tree Least delay path VnVn VpVp s Rsp VnVn s VpVp
24 DCDMR Algorithm Lemma –Let be any non-leaf node on T, is any child of on tree. If then
25 Outline Motivation Network model and problem definition DCR algorithm DCDMR algorithm Performance evaluation Conclusion
26 Performance Evaluation Compare performance with NAIVE and BSMA –NAIVE constructs dynamic tree only satisfy a given delay constraint –BSMA has best cost performance among all static multicast heuristics
27 Simulation Parameters Random network –Link capacities is 155Mbps –average node degree=4 –Equivalent BW of background traffic on each link uniformly distributed between 10 and 120 Mbps –Random generated source s –Random series requests –Link costs Total currently reserved BW on link –Propagation speed on links is two thirds of light Queuing component is neglected
28 First Case of Simulation Every in is add request –For DCDMR and NAIVE Generate new delay-constrained tree when join –For BSMA Directly generate T m for m members without considering request sequence –For simulation results (1) and (2) Delay bound is 50 ms
29 Simulation Results (1)
30 Simulation Results (2)
31 Second Case of Simulation Request may be add or delete is considered How to determine next request? –Consider function : n : number of network nodes k : number of group nodes on current tree : 0.15 r : random number (0<r<1) : new request is delete or add if
32 Simulation Results (3)
33 Simulation Results (4)
34 Conclusion Characteristics of DCDMR –Scales well Source needs small computation –Good cost performance In FAST mode, route computation is fast In SLOW mode, low cost tree is computed