Designing%20Overlay%20Multicast%20Networks%20for%20Streaming

1
Designing Overlay Multicast Networks for Streaming

• Jevan Saks
• Bruce Maggs
• Konstantin Andreev
• Adam Meyerson

2
The Problem

• Delivering streaming media to Media Servers
• Encoder ? Media Server ? User
• Centralized delivery model
• Need scalable and reliable solution

3
The Problem

• Server bottlenecks
• Points of failure
• Can only serve about 50Mbps of streams
• Network bottlenecks
• Unpredictable topology
• Best-effort delivery No guarantees

4
Delivering Streaming Media

• Server bottlenecks
• Points of failure
• Can only serve about 50Mbps of streams
• Network bottlenecks
• Unpredictable topology
• Best-effort delivery No guarantees

5
Live Streaming
Reflectors
Entry Point
Encoder
Edge Servers (Sinks)
6
Proposed Solution

• A less centralized approach
• Entry Point
• Source of the stream, sends to reflectors
• Reflector
• Intermediate server, can split and retransmit
• Edge Server
• Reconstructs a better quality stream from what it
  receives and serves the end user

7
Approach

• Three-tier structure

S Sources
R - Reflectors
D - Sinks
8
Considerations

• Cost
• Capacity
• Bandwidth
• Quality
• Reliability

9
IP Parameters

• Success requirements (?)
• Failure probabilities (p)
• Cost on edges (c)
• Cost on reflectors (r)
• Fanout restrictions (F)

10
Integer Program

• Indicator variables
• zi reflector i used
• yki stream k sent to reflector i
• xkij stream k sent to sink j through reflector i

11
Constraints

• Minimize
• Subject To

12
Solving the IP

• ILOG Cplex (concert library)
• DashOptimization Xpress-MP
• GLPK (GNU Linear Programming Kit)

13
An Example
(Thanks to graphviz)
14
Approximation

• Why? IP is slow, and topology is large and
  changes quickly
• Randomized rounding
• Cost violated by factor O(log n)
• Fanout/weight constraints violated by factor of
  at most 2
• Modified GAP Assignment
• Uses max-flow on a graph of size O(RD)
• Cost violated again by O(log n)
• Fanout/weight constraints violated by another
  factor of at most 2

15
Comparison
Approx Solution (Cost 52)
IP Solution (Cost 51)
16
In Progress

• Multi-source approximations
• Timing comparisons
• Violations in average-case scenarios

17
Extensions

• Capacities on all edges
• Capacities from reflector-edgeserver
• Bandwidth requirements
• Color constraints

18
Extensions
Reflectors
Entry Point
UU Net
Encoder
Edge Server (Sink)
BMM Net
19
Extensions Constraints

• Color constraints
• where RR1 R2 Rm
• (Ri is the set of reflectors on the ith ISP)
• Here different colors represent different ISPs
• There is no added value in serving to a fixed
  sink from two reflectors of the same color.

20
Future Work

• Use real-world data from Akamai
• Implement extensions
• Improve approximation?

21
Best Case Scenario?