Presentation is loading. Please wait.

Presentation is loading. Please wait.

CS5248 Student Presentation1 Scalable Resilient Media Streaming Suman Banerjee, Seungjoon Lee, Ryan Braud, Bobby Bhattacharjee, Aravind Srinivasan NOSSDAV.

Similar presentations


Presentation on theme: "CS5248 Student Presentation1 Scalable Resilient Media Streaming Suman Banerjee, Seungjoon Lee, Ryan Braud, Bobby Bhattacharjee, Aravind Srinivasan NOSSDAV."— Presentation transcript:

1 CS5248 Student Presentation1 Scalable Resilient Media Streaming Suman Banerjee, Seungjoon Lee, Ryan Braud, Bobby Bhattacharjee, Aravind Srinivasan NOSSDAV 2004

2 CS5248 Student Presentation2 Application Layer Multicast Advantages No change in network infrastructure Applications have full control Disadvantages Stretch and Stress Control data overhead Multicast forwarding at end-hosts Construct an overlay network

3 CS5248 Student Presentation3 Examples Narada Builds a mesh, then a tree Everybody knows everybody High control overhead NICE Hierarchical clustering of nodes Low control overhead

4 CS5248 Student Presentation4 NICE

5 CS5248 Student Presentation5 Problem Overlay network node failures Overlay network link failures Congestion failures

6 CS5248 Student Presentation6 SRMS Architecture MS AR B Y X Media Stream Join Request Address of Sender Request Data Data Streaming Server SRMS sender SRMS-RP SRMS client

7 CS5248 Student Presentation7 Probabilistic Resilient Multicast (PRM) Randomized Forwarding Triggered NAKs

8 CS5248 Student Presentation8 Randomized Forwarding Each node in the overlay network forwards the data to a constant number of other overlay nodes with a low probability (0.01–0.03)

9 CS5248 Student Presentation9 Randomized Forwarding (cont’d) A B D C FE NML GH QPOKJI X X

10 CS5248 Student Presentation10 Overhead Analysis n : Total number of nodes r : Number of randomly forwarded nodes β : Probability of random forwarding Per-node overhead of PRM: βr

11 CS5248 Student Presentation11 Triggered NAKs Data losses due to link errors and network congestion are recovered using NAK- based retransmissions using the missing sequence numbers.

12 CS5248 Student Presentation12 Triggered NAKs (cont’d) Each node piggybacks a bit mask with every forwarded packet indicating the prior sequence numbers it has correctly received Recipient of the data packet detects missing packets using the gaps in the received sequence and requests appropriate retransmissions

13 CS5248 Student Presentation13 Triggered NAKs (cont’d) 0111 0011 0111 X Y Z SEQ: 18 NAK: 14, 15 NAK: 16 0000 0011 17 16 15 14

14 CS5248 Student Presentation14 Experiments n : 10 – 10,000 r : 1 - 3 β : 0.01 – 0.03 Compared PRM with Best-Effort (BE) methods Nomenclature: PRM b (r, β) b – bit mask used in NAK retransmissions

15 CS5248 Student Presentation15 Evaluations: Delivery Ratio

16 CS5248 Student Presentation16 Evaluations: Data Loss

17 CS5248 Student Presentation17 Evaluations: End-to-End Latency

18 CS5248 Student Presentation18 Conclusions SRMS achieves high data distribution rates even with node and link failures Very low overhead Scales very well

19 CS5248 Student Presentation19 Q&A


Download ppt "CS5248 Student Presentation1 Scalable Resilient Media Streaming Suman Banerjee, Seungjoon Lee, Ryan Braud, Bobby Bhattacharjee, Aravind Srinivasan NOSSDAV."

Similar presentations


Ads by Google