1. An analysis of a router-based loss detection service for active reliable multicast protocols ICON’02, Singapor M. Maimour, C. Pham RESAM UCB Lyon - INRIA RESO ENS Lyon Thursday, August 29th, 2002

2. 2 A (very) quick overview of multicast Sender data Receiver data Sender data Receiver IP multicast RFC 1122 without multicast with multicast

3. multicast! Everybody's talking about multicast! Really annoying ! Why would I need multicast for by the way? multicast! alone multicast!

4. 4 high-speed www video-conferencing video-on-demand interactive TV programs remote archival systems tele-medecine, white board high-performance computing, grids virtual reality, immersion systems distributed interactive simulations/gaming… Challenges for the Internet Think about…

5. 5 The Wild Wild Web important data heterogeneity, link failures, congested routers packet loss, packet drop, bit errors… ?

6. 6 At the routing level management of the group address (IGMP) dynamic nature of the group membership construction of the multicast tree (DVMRP, PIM, CBT…) multicast packet forwarding At the transport level reliability, loss recovery strategies flow control congestion avoidance Multicast difficulties

7. 7 What is the problem of loss recovery? feedback (ACK or NACK) implosion replies/repairs duplications difficult adaptability to dynamic membership changes Design goals reduces recovery latencies reduces the feedback traffic improves recovery isolation Reliable multicast

8. The reliable multicast universe YOID ALMI HBM Application-based RMANP ARM DyRAM Router supported, active networking AER PGM RLC RLM Layered/FEC CIFL FLID Logging server/replier LBRM SRM TRAM RMTP LMS XTP End to End MTP RMF AFDP 10 human years (means much more in computer year)

9. Routers have specific functionalities/services for supporting multicast flows. Active networking goes a step further by opening routers to dynamic code provided by end-users. Open new perspectives for efficient in-network services and rapid deployment. RMANP ARM DyRAM AER PGM Router supported, active networking

10. 10 Ex: Global NACKs suppression NACK4 data4 NACK4 only one NACK is forwarded to the source

11. 11 The case of loss detection Traditionally the loss detection is performed by : the sender (use of ACKs) in sender- initiated protocols the receivers (use of NACKs) in receiver- initiated protocols We want to investigate a router- supported loss detection service

12. 12 Motivations and design choices Enable early packet loss detection (EPLD) to reduce the latency Routers keep track of the packet sequence: 2 variables per multicast session Implemented as an active service, executed by an active router EPLD services are enabled at specific locations in the network

13. 13 "Satellite picture" of the Internet from UREC, http://www.urec.fr

14. 14 Users' accesses offices campus residentials Network Provider metro ring Network Provider PSTN ADSL Cable … Internet Data Center

15. 15 1 st step: Analysis and network model Each node is modeled by a M/G/1 queue. The delay analysis is based on the mean waiting time of the system.

16. 16 The methodology The different mean flow rates λ 1, λ 2,... λ n of the node with their respective service requirement X 1, X 2,..., X n. The load ρ at this node can be computed using : ρ = Σλ i E[X i ]. The mean waiting time (P-K formula) : E[W]= Σλ i E[X i 2 ] / 2(1- ρ)

17. 17 Performance comparaison Protocol A : Nacks suppression Subcast Protocol D : Nacks suppression Subcast Early Packet Loss Detection service

18. 18 Numerical results The router position influence Loss detection service as a function of the loss rate Maximum loss rate supported by the system Required processing power at the routers so they are never the bottleneck Load at the different nodes The gain as a function of B (# recv)

19. 19 Router position

20. 20 Router position (cont.)

21. 21 Loss detection service gain

22. 22 Max loss rate supported as a function of the processing power of the routers

23. 23 Required processing power at the routers so they are never the bottleneck

24. 24 Load at the different nodes low overhead!

25. 25 The gain as a function of B

26. 26 2 nd step: Adding EPLD in DyRAM DyRAM is an active reliable multicast protocol with local recoveries from elected repliers more accurate model

27. 27 Simulation results p=0.25 #grp: 6…24 4 receivers/group EPLD is very beneficial to DyRAM simulation results very close to those of the analytical study

28. 28 Conclusions Early packet loss detection by routers is found to enhance the performances of reliable multicast Reduction of the recovery latency is targeted to enable distributed applications on computational grids Simulations and experimental test-beds are encouraging Incorporated into egde-routers?