1. Multi-Source Latency Variation Synchronization for Collaborative Applications Abhishek Bhattacharya, Zhenyu Yang & Deng Pan

2. Roadmap Introduction Problem Motivation Heuristic Solution Results Summary

3. Introduction Single-source/Single-stream Multicast Network: Construction of a MST connecting a single source and multiple receiver nodes Single-source/Multi-stream Multicast Network: Content distribution systems using MDC/SVC streams. Multi-source/Single-stream Multicast: Constructing a forest of trees connecting multiple sources with multiple receiver nodes. Multi-source/Multi-stream systems such as 3D Virtual Immersive Systems, 3D Videoconferencing, Online games, etc.

4. Example – Multi-source/Multi-stream System 3DTI environment provides a collaborative virtual space for geographically distributed users. Multiple 3D cameras installed for capturing the same scene from various viewpoints. Each 3D camera produces a video stream. Multiple streams are transmitted from each node to all other nodes since each node acts a source and a receiver.

5. Node A Node B Node C 3D Camera Display Unit S 1 AB S 2 AB S 7 BA S 6 BA S 8 CB S 1 CB S 5 BC S 4 BC S 3 AC S 7 CA S 6 CA S k ij camera stream ‘k’ from node ‘i’ to node ‘j’

6. Introduction Latency Variation is an important QoS constraint for Multi- source/Multi-stream systems. Problem known as Delay Variation Bounded Multicast Network (DVBMN) in the literature. Proved to be NP-complete by Rouskas et al. and thereafter many heuristics proposed. Multi-source/Multi-stream latency and latency-variation constraints: E2E from source to destination, Inter-stream, Inter- Source & Intra-Source Latency Variation. Intra-source is important for 3DTI applications due to the high correlation factor among the multiple streams from the same source.

7. Node A Node B Node C S 1 AB S 2 AB S 7 BA S 6 BA S 8 CB S 1 CB S 5 BC S 4 BC S 3 AC S 7 CA S 6 CA Intra-SourceVariation Inter-Source Variation

8. Problem P:P: C1:C1: C2:C2: C3:C3: E2E Latency from Source to Destination: Inter-Stream Latency Variation: Inter-Source Latency Variation:

9. Motivation V1V1 V3V3 V2V2 V4V4 V6V6 V7V7 V5V5 6 5 8 12 10 2 12 7 4 8 9 ∆ = 22 S 1 6 : V 1  V 4  V 6 : 12 S 1 7 : V 1  V 4  V 7 : 16 S 2 6 : V 2  V 4  V 6 : 16 S 2 7 : V 2  V 5  V 7 : 12 β = 16 – 12 = 4 λ 6 = 16 – 12 = 4 ; λ 7 = 16 – 12 = 4 λ T = Max(λ 6, λ 7 ) = 4 V 1  V 3  V 4  V 6 : 17 V 1  V 3  V 4  V 7 : 21 V 2  V 1  V 4  V 6 : 17 V 2  V 4  V 7 : 20 β = 21 – 17 = 4 λ 6 = 17 – 17 = 0 ; λ 7 = 21 – 20 = 1 λ T = Max(λ 6, λ 7 ) = 1

10. Motivation Path Latency List K-shortest Paths Latency S 1 6 V 1  V 4  V 6 12 V 1  V 3  V 4  V 6 17 V 1  V 3  V 6 18 S 1 7 V 1  V 4  V 7 16 V 1  V 2  V 5  V 7 17 V 1  V 3  V 4  V 7 21 S 2 6 V 2  V 4  V 6 16 V 2  V 1  V 4  V 6 17 V 2  V 5  V 7  V 6 21 S 2 7 V 2  V 5  V 7 12 V 2  V 4  V 7 20 V 2  V 1  V 4  V 7 21

11. Heuristic Solution: Algorithm  A 2-step framework:  K-shortest-path Algorithm:  Involves the computation of k-shortest path from each source to all the destination nodes.  Generates ‘d*s’ lists with ‘k’ elements in each list.  State-of-art ksp Algorithm: Recursive Enumeration Algorithm(REA) by Jimenez et al.

12. Heuristic Solution: Algorithm  M SLV Algorithm:  Involve a selection of paths for the construction of forest i.e., multiple multicast trees connecting one source node to the set of destination nodes.  Selecting ‘d*s’ path latency values from ‘d*s*k’ elements. Path Latency values.  The Path Latency values should satisfy the constraints: C 1 : E2E latency bound (∆), C 2 : Latency Variation among any path(β), and C 3 : Inter-Source LatencyVariation(λ T ).

13. Heuristic Solution: Algorithm 1718  T = 4; λ 6 = 4; λ 7 = 4; λ T = 4 12 S16S16 S26S26 S17S17 S27S27 172116 202412 172116 12 16 12 16 1712  T = 5; λ 6 = 1; λ 7 = 4; λ T = 4 2012  T = 4; λ 6 = 1; λ 7 = 4; λ T = 4 1617  T = 4; λ 6 = 1; λ 7 = 3; λ T = 3 1617  T = 3; λ 6 = 0; λ 7 = 3; λ T = 3 1718  T = 4; λ 6 = 0; λ 7 = 1; λ T = 1 1721  T = 4; λ 6 = 4; λ 7 = 1; λ T = 4 1721  T = 3; λ 6 = 3; λ 7 = 1; λ T = 3 ∆ = 25 ; β = 4

14. Heuristic Solution: Time Complexity  ksp-Algorithm: O(m + nk * log(m/n)) m  number of total edges in the network graph n  number of total nodes in the network graph k  number of shortest paths  MSLV Algorithm: O(dsk * log(ds)) d  number of destination nodes in the multicast set s  number of source nodes in the multicast set

15. Results Detailed results in the paper

16. Summary Studied construction of multicast networks with multiple sources and receivers. Satisfying different Latency Variation constraints which are important for real-time multi-party/multi-stream systems. A 2-step heuristic framework consisting of an initial ksp-algorithm to generate shortest paths from sources to receivers followed by a path selection process to satisfy the various hard/soft constraints. Future work involves to investigate the problem with link capacities as time-varying functions and decentralized solutions with node/network dynamics.