Layered Peer-to-Peer Streaming

Yi Cui, Klara NahrstedtDepartment of Computer Science

University of Illinois at Urbana-ChampaignSource International Workshop on NOSSDAV’03, June, 2003

Outline

Problem addressingAsynchronousHeterogeneity

Layered peer-to-peer streaming solutionUnlimited number of supplying peersConstraint supplying peersLayered rate heterogeneity

Performance evaluation Conclusion

Problem addressing

AsynchronousUser request media data at different timeSolution: cache-and-relay approach

HeterogeneityRequest stream of different qualities due to

resource constraints such as network bandwidth

Solution: layer-encoded streaming approach

Problem addressing (cont.)

Layered peer-to-peer streaming

FeatureLimited inbound/outbound

bandwidth Goal

Maximize total qualities

Subject to

Qk: total receive layers of peer k

Ik: inbound bandwidth

Ok:outbound bandwidth

Hk:peers

Layered peer-to-peer streaming

lk: inbound bandwidth of Hk (# of layers) Ak: available layers at the cache of Hk

H0: server,

S = { H1, H2, …, HM } : set of hosts sorted by available layer number

i.e., A1≤ A2 ≤ … ≤ AM

Qkm: # of layers get from host m

Qk: streaming quality

Basic algorithm (cont.)

Basic algorithm

Available cache layers Outbound bandwidth

H1

H2

H3

H4

Hk

Get from server !

Allocate maximum # of layers for Hk

Enhanced algorithm

Ck: constraint on maximum # of supplying peers Qk*(M, Ck): optimal solution if Hk can only choose

Ck supplying peers from H1~HM

Qmax(Hm+1, …, HM):

best contributor in { Hm+1, …, HM } Maximize total receive data at Hk

DP implementation ─ O(CkM2)

Fault tolerance

Normal departureDue to user logout or quality degradationThe departure peer notifies Hk to reconfigure

FailDue to machine crash or network failureEither temporally request from server

or suffer from quality degradation

Fault tolerance (cont.)

Available cache layers Outbound bandwidth

H1

H2

H3

H4

Hk Access from server or degradation !

Layered rate heterogeneity

Layer rate allocation schemesNatural Number Scheme

l0 = r0, rk = k r‧ 0

Exponential Scheme rk = r0 2‧ k

Fibonacci Scheme r1 = 2r0, rk = rk-1 + rk-2

Layered rate heterogeneity (cont.) ri: streaming rate of layer I (Kbps)

Ik, Ok: inbound and outbound bandwidth (Kbps)

Performance evaluation

40000 peersModem/ISDN: 50%, 112Kbps (max) Cable Modem/DSL: 35%, 1Mbps (max)Ethernet peers: 15%, 10Mbps (max)

60-min video, which consists of 50 layers, with full quality streaming rate = 1Mbps

Run 24 hours

Overall streaming quality and scalability Streaming quality satisfaction = Qk/Ik

Tradeoff between overall quality & constrained supplying peers

Fairness

Robustness

Layer rate heterogeneity

Conclusion

Introduce a layered peer-to-peer streaming approach to optimize the streaming quality of heterogeneous peers, save server bandwidth.

Hope to make best use of bandwidth resource of supplying peers.

Evaluate the solution by: Test fairness among peers according to streaming

quality satisfaction and bandwidth contribution. Test robustness against unexpected departures/fails.