DVBMN-l: Delay Variation Bounded Multicast Network with multiple paths

Abhishek Bhattacharya & Zhenyu Yang

Roadmap

• Introduction

• Motivation

• Proposed Heuristic

• Simulation Results

• Conclusion

Introduction

• Distributed multi-party/multi-stream systems such as 3D Tele-immersion, Networked Virtual Environments and Multi-Player Online games are becoming popular

• Multiple streams are streamed to different sites through an overlay network and application-level multicasting

• Synchronization among the different streams is crucial for real-time and collaborative interaction

• QoS guarantees such as end-to-end delay and delay variation to be satisfied

MotivationSource : VS

Multicast Set (M): V2, V5, V8

•End-to-End Delay(∆) refers to the delay between the source and the destination node• Goal is to select paths so that VS,V2, V5 and V8 are connected

• If we only consider only ∆ then the multicasting solution is: Vs V1 V2 (31) Vs V7 V8 V5 (26) Vs V7 V8 (20)

Motivation• Consider delay variation which is the max variation among the

paths from VS to V2, V5 and V8

to minimize: max (|D (s, i) – D (s, j)|)• Optimal result:

Vs V7 V8 V5 V4 V2 (40)

Vs V1 V2 V5 (40)

Vs V1 V2 V4 V8 (43)• Delay Variation = 3 (optimal)

• This is the case when l = 1

• We consider delay variation for multiple paths: 1<= l <= kwhere k is satisfied by all path values less than Δ

Motivation• When l > 1 then there are 2 more constraints to be considered:• Inter-destination delay variation: delay variation between path

delay values to V2, V5 and V8

• Intra-destination delay variation: delay variation between path delay values among k multiple paths of V2 or V5 or V8

• We consider the combined inter-intra destination delay variation among k-paths from VS to V2, V5 and V8

• For l = 2 the solution will be:

Vs V7 V8 V5 V2 (35)

Vs V7 V8 V5 V4 V2 (40)

Vs V1 V2 V5 (40)

Vs V1 V2 V4 V5 (45)

Vs V1 V2 V4 V8 (43)

Vs V1 V2 V5 V8 (46)

Inter-intra delay variation = 11

Proposed Heuristic• A two-step framework

• First stage involves using an efficient k-shortest path algorithm from the literature which returns a path list for every v Є M

• All the entries from the above path list satisfy source-to-end maximum delay bound and are sorted in ascending order

• The second step involves selecting those optimal paths for each destination node such that the max inter-intra delay variation among those paths have the tightest possible bounds

Proposed Heuristic

For k = 4 and Δ = 55 the k-shortest path algorithm will return:V2: 31, 32, 35, 40V5: 26, 32, 40, 45V8: 20, 43, 46, 52

Proposed Heuristic• Initially we extract l elements from each of the three lists and

insert then into a heap

• In each iteration, we calculate the delay variation and update if lesser than the previous one

• Then, we extract the least element from the heap and insert the next element from same destination list into the heap

• This process will exhaustively search all the possible combinations and at the end will keep the best possible combination of path delay values

• Time Complexity: O(mk log m)

Proposed Heuristic• The execution path of the algorithm will be as follows:

• V2: 31, 32, 35, 40

• V5: 26, 32, 40, 45

• V8: 20, 43, 46, 52

• Iteration 1: 20, 26, 31, 32, 32, 43 <23>• Iteration 2: 26, 31, 32, 32, 43, 46 <20>• Iteration 3: 31, 32, 32, 40, 43, 46 <15>• Iteration 4: 32, 32, 35, 40, 43, 46 <14>• Iteration 5: 32, 35, 40, 40, 43, 46 <14>• Iteration 6: 35, 40, 40, 43, 45, 46 <11>

• Optimal solution is: 35, 40, 40, 43, 45, 46 <11>

Simulation Results• We performed experiments comparing the execution times of

Chains Proposed by Banik et. al.

• Result shows notable performance gains of our approach compared to Chains

• Our approach is also scalable since the execution time grows very slowly with the increase in the number of destination nodes unlike Chains which has a sharp rise

Simulation Results

Conclusion• We investigated the multi-stream synchronization problem for

3DTI and other collaborative applications which require strict QoS guarantees such as delay variation bounds with real-time requirements

• The solution is proposed in a two-step framework consisting of finding the k-shortest path and then the optimal path list search algorithms

• Our solution satisfies the tightest inter-intra destination delay variation bound along with the end-to-end delay bound for multiple paths

• Time complexity of our solution is O(mk log m) which is better than Chains {O(m2k) } even for l = 1

• Future work involves considering the synchronization problem with multiple sources, embedding this component into the current 3DTI architecture and various issues related to network and link dynamics