Issues in Offering Live P2P Streaming Service to Residential Users Nazanin Magharei, *Yang Guo, and Reza Rejaie Dept. of Computer and Information Science.

Issues in Offering Live P2P Streaming Service to Residential Users

Nazanin Magharei, *Yang Guo, and Reza Rejaie

Dept. of Computer and Information Science *Princeton CR Lab University of Oregon Thomson Inc.

3

Outline

Introduction and related work PRIME: Mesh-based P2P streaming service Issues in offering p2p streaming to residential users

Effect of available resource Effect of heterogeneous bandwidth Effect of freeloaders Effect of number of users

Conclusions and summary

4

Introduction

P2P technique attracting attentions from commercial world NBC Universal goes peer-to-peer – wurldmedia.com BitTorrent raised $8.75 million venture capitals

Teamed with CacheLogic to work for BT Startups providing P2P live program: pplive, coolstreaming BBC IMP Why?

Reduce the cost to compete with piracy

Conceivably provide p2p live streaming in a commercial setting

Using mesh-based p2p streaming

5

Introduction

P2P live streaming Tree-based approach

ESM, SplitStream, etc. Mesh-based approach

Coolstreaming, Chainsaw, PRIME, etc. Fundamental difference – static mapping of content to

delivery topology vs. dynamic mapping

Pkt delivery time

Bandwidth variation

Peer degree

Group size

Persistent churn

Batch departure

Mesh

Tree

6

Introduction and Related Work

Challenges Heterogeneous access speed – DSL, cable modem, …

Insufficient resource Asymmetric bandwidth – uplink bandwidth < downlink bandwidth Free-loaders

Not willing to contribute Cannot contribute

Behind NAT box or firewall

Key questions What is the impact of available resource to overall performance? How similar (different) is such an effect across peers with different

bandwidth? Whether and how the freeloaders affect the overall performance

and individual received quality?

7

PRIME: Mesh-based P2P Streaming Service

Peer expects to receive maximum deliverable quality through its access link

Using MDC in content delivery Two possible performance bottlenecks

Bandwidth bottleneck Insufficient aggregate bandwidth from all parents

Content bottleneck Insufficient useful content from all parents

PRIME attempts to minimize these bottlenecks

8

Global Pattern of Content Delivery

Connections in the overlay have roughly the same bandwidth

Group peers into levels, based on their shortest distance from source

Each peer with degree d in level n has at least one parent in level n-1 (diffusion parent) and d-1 parents in the same or lower levels (swarming parents)

21 3

6 754

12 13910

118

Source

Level 1

Level 2

Level 3

depth

9

Global Pattern of Content Delivery

Diffusion phase Peers should receive a

data unit as fast as possible

Swarming phase Peers exchange

(swarm) data units with each other until receive their desired quality of the segment

Level 2

Level 3

Level 1 21 3

6 754

12 13910

118

SRC

10

Simulation Setting

Evaluated using ns with congestion control Network topology generated using Brite Video rate of 400 kbps, downlink bandwidth of 550 kbps Various resource distribution

Uplin Bw SC1 SC2 SC3 SC4 SC5 SC6

128kpbs 27% 54% 13% 5% 11% 50%

384kpbs 60% 20% 80% 9% 14% 39%

1Mbps 13% 26% 7% 36% 25% 11%

0kbps 0% 0% 0% 50% 50% 0%

RI 1 1 1 1 0.8 0.8

11

Effect of Available Resource

Resource Index

Avg

. re

ceiv

ed q

ualit

y

CDF of received quality

Average received quality is proportional to the resource index, however the individual received quality is random

12

Effect of Heterogeneous Bandwidth

Avg

. re

ceiv

ed q

ualit

yCDF of received quality

•Bandwidth heterogeneity has no impact on the peers’ received quality•No correlation between received quality and resource contribution

Upload bandwidth

13

Effect of Free-loaders

Free-loaders degrade the connectivity between different diffusion trees, hence prevent content swarming and limit delivery quality

14

Effect of Number of Users

15

Summary

Two issues identified In resource poor scenarios, the delivered quality to

peers is not correlated to their contribution P2P streaming can handle heterogeneous

bandwidth, however the presence of free-loaders significantly affect the mesh connectivity and degrade delivered quality

Solution: contribution-aware p2p streaming Delivered quality is proportional to contribution Encourage cooperation

Backup Slides

17

Global Pattern of Content Delivery

Level 2

Level 3

Level 1 21 3

6 754

12 13910

118

SRC

18

PRIME: Mesh-based P2P Streaming Service

Prior studies often assume a fix peer degree Bandwidth bottleneck only depends on overlay

topology Incoming/outgoing bandwidth of participating peers Incoming/outgoing degree of participating peers

Avg. BW for a connection between parent p and child c

MIN (outbwp/outdegp, inbwc/indegc) All connections in the overlay have roughly the

same bandwidth