Performance in Decentralized Filesharing Networks Theodore Hong Freenet Project.

Performance in Decentralized Filesharing Networks

Theodore HongFreenet Project

2

Styles of collaboration

Centralized model e.g. Napster global index held by central authority

(single point of failure) direct contact between requestors and

providers

Decentralized model e.g. Freenet, Gnutella no global index – local knowledge only

(approximate answers) contact mediated by chain of intermediaries

3

Key questions

Does it work? can we find the data? query success rates

length of query paths

Does it scale? logarithmic / linear / polynomial

Is it robust? participants are unreliable different failure modes possible

4

An abstract model

Can model the network as a graph:

5

Querying the network

Answering a query means finding a path source = requestor destination = provider

A distributed search problem! approximate global solution using local

knowledge same problem as IP routing

6

The Freenet algorithm

Graph structure actively evolves over time new links form between nodes files migrate through network adaptive routing

7

Initial simulations

Ring topology, 1000 nodes:

8

Initial simulations (cont’d)

9

Why does it work?

The small-world model Milgram: six degrees of separation Watts: between order and randomness

short-distance clustering + long-distance shortcuts

10

Links in the small world

P(n) ~ 1/n 1.5

“Scale-free” link distribution P(n) = 1 /n k

most nodes have only a few connections some have a lot of links

11

Small-world links (cont’d)

Real-world examples movie actors (Kevin Bacon game) world-wide web nervous system of worm C. elegans

12

The importance of routing

Existence of short paths is not enough – they must be found

Adaptivity helps Freenet find good paths

Compare: a random-routing network

13

Scalability

Real-world networks are much larger nearly 400,000 downloads of Freenet 50 million Napster users

How well does Freenet scale?

14

Fault-tolerance

Unreliability is normal in peer-to-peer

Two types of failure: random failure targeted attack

15

Random failure

16

Targeted attack

17

To do

Variable disk/bandwidth capacity if you build it, will they come?

Participants leaving and re-entering

File lifetimes “lifetime” is relative relationship between ease of retrieval and

popularity, size impact of splitting and combining

18

Conclusions

Local approximations can be good enough

Small-world model provides useful framework

Metrics to consider: query pathlength clustering coefficient link distribution

Issues to consider: scalability fault tolerance under various scenarios

19

For more information

“Performance” chapter in Peer-to-Peer

I. Clarke, O. Sandberg, B. Wiley, T.W. Hong, “Freenet: a distributed anonymous information storage and retrieval system,” in Workshop on Design Issues in Anonymity and Unobservability, ed. by H. Federrath. Springer: New York (2001)