Placement of Continuous Media in Wireless Peer-to-Peer Network Shahramram Ghandeharizadeh, Bhaskar Krishnamachari, and Shanshan Song IEEE Transactions.

Placement of Continuous Media in Wireless Peer-to-Peer Network

Shahramram Ghandeharizadeh, Bhaskar Krishnamachari, and Shanshan SongIEEE Transactions on Multimedia, April 2004

H2O Framework

Home-to-Home Online (H2O) devices collaborate to deliver continuous media

H2O may act as: A producer of data An active client A router

Motivation

A new replication technique that Provide on-demand access to continuous

media Minimize the total storage space

required

Assumptions

CBR continuous data Total size of available clips exceeds

the storage capacity of one device Bandwidth between two H2O devices

exceeds the bandwidth required to display a clip

One hop distance is a constant

Hi: the Farthest Number of Hops a Block Can be Located

Cycle: period to display a block D=Sb/BDisplay

The farthest number of hops that the block i can be located: Hi=((i-1)D)/h

block size playback rate

time to retrieve a block from one hop away

Data Placement and Replication For each video clip X:

Divide X into equal-sized blocks with size Sb Place first block, b1 on each node. For each block bi, 1<i<=z, compute delay toler

ance Hi Compute ri based on Hi Construct ri replicas of bi and place them

ri is a topology dependent computation

Topology I: Worst Case Linear Topology

Block i should be replicated ri times: Hi=(i-1)D/h ri=N-Hi Reset ri to one if ri is zero or negative

Total storage space (SC,R) occupied by a clip with z blocks:

1 2 3 8 9…

z

i

z

i ibibRC rSrSS1 1, )(

Percentage Saving Compared with Full Replication in Linear Topology

•N=1000, h=0.5,•BDisplay = 4Mbps•y: 100x(1-SC,R)/(SCxN)

Topology II: Grid Topology

Organize N nodes in a square area At least one copy of bi must be placed

within Hi hops There are nodes within Hi

hops of every node

Total storage required:

122 2 ii HH

122 2

iii HH

Nr

z

iii

b

z

i biRC HH

NSSrS

1 21, 122

Total Storage Space Required as a Function of Block Size (1/2)

•h=0.75s

•2 min clip (total 60MB)

Total Storage Space Required as a Function of Block Size (2/2)

•h=0.75s

•2 hour clip (total 3600MB)

Topology III: Average Case Topology (1/2)

Network connectivity depends on radio range R

N nodes are scattered in area A There are on average between and

nodes within Hi nodes.)/())(2( 2 ANRH i )/())(( 2 ANRH i

Topology III: Average Case Topology (2/2)

Using the upper boundary, the H number of replicas ri required by bi is:

Total storage required for a clip:S

2)( RH

Ar

ii

z

i

z

i ibibRC rSrSS1 1, )(

Percentage Saving Comparison

Distributed Implementation H2Op: publish a clip X

Compute block size Sb, number of blocks z, and Hi for each block

Flood the network to query which H2O will host a copy of which block of X

H2Oj: each recipient of the message Compute a binary array Aj that consists of z e

lements whose values are 0 or 1 Two computation methods: TIMER or ZONE

Technique I: TIMER

When H2Oj receives query message Perform z rounds of elections Pick a random timer value between 1

and M then count down The one first count down to zero stores a

copy and send suppress message within Hi hops

May generate more than one copies of a block within Hi hops

Technique II: ZONE

Assume each node is aware of its (x, y) coordinate

Place each copy in a separate square zone whose size is such that all nodes can be reached within Hi hops

Simulation: TIMER vs. ZONE

•N=300, R=100m, A=1km2, z=60

Simulation: Comparison of Analytical Models for Graph Topology with 2 Implementations

SC=60MB R=100m A=1km2

Simulation: How Many Blocks a H2O Device Have When Using TIMER

•N=300, R=100m, A=1km2

•Average # of blocks per node for a clip is marked as dashed line

Conclusion

Provide a novel replication technique for on-demand clips Minimize startup delay Storage saving compared with full

replication Provide two distributed

implementations