PROMISE: PeertoPeer Media Streaming Using CollectCast

1
PROMISE Peer-to-Peer Media Streaming Using
CollectCast

• Mohamed Hefeeda, Ahsan Habib, Boyan Botev,
  Dongyan Xu, Bharat Bhargava
• Department of Computer Sciences
• Purdue University, West Lafayette, IN 47907

2
Outline

• Abstract
• Introduction
• PROMISE An overview
• Select best peers
• Rate and data assignment
• Dynamic switching
• Evaluation/Internet experiments
• Conclusion and future work

3
Abstract

• We present PROMISE
• A novel peer-to-peer media streaming system
• Design
• Implementation
• Evaluation
• Key functions of PROMISE
• Peer lookup
• Peer-based aggregated streaming
• Dynamic adaptations to network and peer conditions

4
Abstract cont.

• PROMISE is based on a new application level P2P
  service called CollectCast
• Selection of sender
• Inferring and leveraging the underlying network
  topology and performance information
• Monitoring the status of peers and connections
• Reacting to peer/connection failure or
  degradation with low overhead
• Dynamically switching
• active senders
• standby senders

5
Introduction

• In a P2P system, peers communicate directly with
  each other for the sharing and exchange of
• Data
• Storage
• CPU capacity

6
Introduction cont.

• This paper focus on
• P2P real-time media streaming
• Involve multiple sending peers in one streaming
  session

7
Introduction cont.

• The dynamic and diversity are reflected in both
  peers and network connections between peers
• A sender may stop contributing at any time
• The outbound bandwidth contributed by a sender
  may change
• The connection may exhibit different end-to-end
  bandwidth, loss, and failure rate
• The underlying network topology that determines
  the connections are not independent of each other

8
PROMISE An overview

• Consists of a set of peers interconnected through
  a P2P substrate
• Operations are independent of the underlying P2P
  substrate
• Modified Pastry to return multiple peers for each
  lookup request

9
PROMISE architecture
10
PROMISE Notations

• Ap availability
• 1.0 available for streaming
• 0.0 - otherwise
• R0 playback rate

11
CollectCast and PROMISE

• The design of PROMISE relies on a novel
  application level P2P service called CollectCast
• The receiver plays the leading role in
  CollectCast
• Selecting the best sending peers
• Inferring and monitoring the underlying network
• Assigning streaming rate and data segment to the
  sending peers
• Deciding when a change of the sending peers is
  needed

12
PROMISE Operation

• Two connections are established with each peer
• A UDP connection for sending the stream packets
• A TCP connection for sending control packets

13
CollectCast Receiver side
14
CollectCast Sender side
15
Selecting best peers

• Three selection techniques are possible
• Random
• End-to-end
• Topology-aware

16
Random Selection

• Randomly chooses a number of peers that can
  full-fill the aggregate rate requirement
• these peers may have low availability and share a
  congested path

17
End-to-end Selection

• Estimates the goodness of the path from each
  candidate peer to the receiver
• Based on the quality of the individual paths and
  on the availability of each other

18
End-to-end Selection cont.
19
End-to-end Selection cont.
20
End-to-end Selection cont.

• Ex.
• P3, P5, P6
• 1 .8 1 .8 1 .9 2.5
• The best one
• P3, P5, P6 and P1, P2, P3, P6
• 2.5

21
Topology-aware selection

• Infers the underlying topology and its
  characteristics
• Considers the goodness of each segment of the
  path
• Make a judicious selection by avoiding peers
  whose paths are sharing a tight segment

22
Topology-aware selection cont.
23
Topology-aware selection cont.
24
Topology-aware selection cont.

• Ex.
• P3, P5, P6
• 1 .8 1 .8 .25/.50 .9 2.05
• The best one
• P2, P3, P6
• 2.4

25
Selection Algorithm
26
Rate and Data Assignment

• Forward Error Correction (FEC)
• Use erasure codes to tolerate packet losses
• FEC(a)
• The system can tolerate up to (a-1) packet loss
  rate
• A segment of ? packets
• encoded into ? /(2-a)

27
Rate and Data Assignment cont.
28
Rate and Data Assignment cont.

• Ex.
• RP1 R0/2, RP2 R0/4, RP3 R0/2
• al 1.0625, au 1.25, a 1.125
• Result.
• RP1 .45, RP2 .225, P3 .45
• DP1 55, DP2 38, DP3 55

29
Dynamic Switching

• Peer failure detected
• From the TCP control channel (eq. connection
  reset)
• The rate coming from this peer is degraded
• Actions
• Assigning new rates for he currently active peer
  set
• Adjusting the active set by adding or replacing
  peers

30
Evaluation

• R0 1Mb/s
• Streaming session lasts for 60 mins
• AP .5, .9
• RP .125R0, .5R0

31
Aggregated loss rateno peer failure
32
Aggregated streaming rate no peer failure
33
Aggregated loss ratewith peer failure
34
Aggregated streaming ratewith peer failure
35
Size of the candidate peers set required /
availability values
36
Conclusion

• The most salient features of PROMISE
• Accounts for peer heterogeneity, reliability, and
  limited capacity
• Matches a requesting peer with a set of supplying
  peers that are likely to achieve the best
  streaming quality
• Dynamically adapts to network fluctuations and
  peer failure
• Inferring and leveraging the underlying network
  conditions