1. 1 Enabling Contribution Awareness in an Overlay Broadcasting System ACM SIGCOMM 2006 Presented by He Yuan

2. 2 Outline Background Related Work Contribution-aware Design Implementation and Experiments Conclusion Discussion

3. 3 Video Broadcast using Overlay Multicast Tokyo LA San Francisco Boston Pisa Encoder E E D D E D D: DSL E: Ethernet Pisa Tokyo NYC LA Boston San Francisco Overlay Tree

4. 4 Background I State-of-Art in Overlay Broadcast Architecture and Protocol Design Narada, SplitStream, CoopNet, DONet... Significant progress in scalability & resiliency Real Deployments ESM *, CoolStreaming, PPLive, SopCast...

5. 5 Background II Much success to date: Homogeneous environments with abundant bandwidth Heterogeneity in node upload bandwidth Upload access bandwidth varies widely Hosts may choose to forward differently Insufficient bandwidth resource DownloadUpload DSL600-1200Kbps64-256Kbps Cable1-6Mbps128-768Kbps Ethernet≥ 10Mbps > 80% < 20%

6. 6 Related Work Bit-for-bit policy Effective only in BT-like systems Differential Admission Control Not feasible in the mainstream Internet Taxation model Incentive vs. Fairness

7. 7 Goals and Challenges Goals Good utilization of bandwidth Differential and equitable distribution Guaranteed QoS Challenges More generic than bit-for-bit policy Distributed sampling and computing Dynamic environment

8. 8 Contribution-aware Design Assumptions Multi-tree-based data dissemination Bandwidth distribution policy System design

9. 9 Assumptions Abundant download bandwidth Different levels of contribution Actual contribution f i reflected by Forwarding bound F i Non-strategic honest clients To encourage a host to relax its F i

10. 10 Multi-tree-based data dissemination Using MDC, split into T-equally sized stripes T trees, each distributes a single stripe of size S/T Overall quality depends on the number of stripes received Number of trees node i is entitled to = Source S/3 S Kbps Tree 1 Tree 3 Tree 2 Peer A Peer C

11. 11 Bandwidth distribution policy Entitled bandwidth 0 < α < 1 Contribution ∑ f j / N j More generic than bit-for-bit Differential and Equitable Distribution

12. 12 Bandwidth distribution: Example Source E E E E D D 100Kbps S = 400Kbps T = 4 avg f = 300Kbps α = 0.5 f E = 500Kbps f D = 100Kbps –r E = 0.5*500+0.5*300 = 400Kbps  entitled to 4 trees –r D = 0.5*100+0.5*300 = 200Kbps  entitled to 2 trees DD Entitled Node Excess Node

13. 13 System Design Distributed System Sampling Computing Number of Entitled Trees Smoothing Locating Excess Bandwidth Backoff in Excess Tree Contribution-Aware Node Prioritization

14. 14 Implementation and Experiments Use Slashdot to evaluate 2 systems: Cont-Agnostic: multi-tree broadcast system Cont-Aware: multi-tree + contribution-aware heuristics S=400Kbps, T=4, stripe size S/T=100Kbps 2 types of peers: Ethernet fmax ≤800Kbps, DSL fmax ≤100Kbps HC: 700-800Kbps, LC: 75-100Kbps Mainstream Internet Conferences Broadcast Event DSL (100Kbps ) Ethernet (10Mbps ) Peak Group Size SIGCOMM200248%52%78 SOSP200348%52%54 Rally75%25%481 Slashdot73%27%158 GrandChallenge82%18%276

15. 15 Evaluation Goals Fairness Overall quality of playback Stability

16. 16 Performance: High Contributors SystemMeanStd. Dev Cont-Agnostic35360.9 Cont-Aware41524.6 Better Cont-Aware gives HC better performance

17. 17 Performance: Low Contributors SystemMeanStd. Dev Cont-Agnostic31180.5 Cont-Aware29534.8 Similar performance among similar contributors Better

18. 18 Stability Time between Tree Reductions Cont-Aware performs slightly worse Reductions => slight dips in quality Not complete disconnection, 63.4% from 4  3, 34.1% from 3  2, only 2.5% from 2  1 and 1  0 Reconnection time (in sec) Cont-AwareCont-Agnostic HC7.180.82 LC53.4265.26 Overall48.2569.83

19. 19 Performance across traces for high contributors

20. 20 Conclusion Resource-scarce, heterogeneous environments Two key ideas: Multi-trees and Linear Taxation Provide fairness in overlay broadcasting in mainstream Internet environments

21. 21 Discussion Applying MDC to Multi-tree overlay The issue of redundancy in coding What’s different in the resulting system? More bandwidth resource or Better QoS Incentive or fairness Where to go? Customized user requirement - Demand according to capacity Location-aware streaming reuse technique

22. 22 Thanks!