[PPT] - Fixed random network Ozalp Babaoglu Average over 500 broadcasts x PowerPoint Presentation

SLIDE 1

Inducing Cooperation in Peer-to-Peer Systems

Ozalp Babaoglu Dipartimento di Scienze dell’Informazione Università di Bologna Italy

University of Bologna

Ozalp Babaoglu

Outline

P2P networks are usually open systems
Possibility to free-ride
High levels of free-riding can seriously degrade global

performance

We present a simple algorithm (SLAC) that sustains

high levels of cooperation despite selfish nodes

We show that certain types of cheating and lying

behavior do not necessarily destroy cooperation (on the contrary, may even improve it!)

2

University of Bologna

Ozalp Babaoglu

The SLAC Algorithm

Node p periodically executes the following:

q = SelectPeer()

if utilityq > utilityp drop all current links link to node q and copy its strategy and links mutate (with low probability) strategy and links fi

Peer selection based on a random overlay network (newscast), whereas copying, rewiring and mutating are with respect to an application (strategy) over an “interaction network”

3

University of Bologna

Ozalp Babaoglu

SLAC Algorithm: “Copy and Rewire”

4

B C A D E F H J K G

C

m

p a r e u t i l i t i e s

“Rewire” A “Copy” strategy

SLIDE 2

University of Bologna

Ozalp Babaoglu

A “Mutate” strategy

SLAC Algorithm: “Mutate”

5

B C D E F H J K G Drop current links Link to random node

University of Bologna

Ozalp Babaoglu

Prisoner’s Dilemma

We test SLAC with Prisoner’s Dilemma (PD)
Captures the conflict between “individual rationality” and

“common good”

Defection (D) leads to higher individual utility
Cooperation (C) leads to higher global utility
T > R > P > S and 2R > T+S

6

C D C R,R S,T D T,S P,P

University of Bologna

Ozalp Babaoglu

Prisoner’s Dilemma

Prisoner’s Dilemma in SLAC
Nodes play PD with neighbors chosen randomly in the

interaction network

Only pure strategies (always C or always D)
Strategy mutation: flip current strategy
Utility: average payoff achieved

7

University of Bologna

Ozalp Babaoglu

Why Does SLAC Work?

SLAC produces very high levels of cooperation
Nodes “move” throughout the network to find better

neighborhoods

This results in an evolution of the (interaction) network
Group-like selection between clusters
Clusters of cooperating nodes grow and persist
Defecting nodes tend to become isolated

8

SLIDE 3

University of Bologna

Ozalp Babaoglu

Example

500 nodes
Initial state:
All defectors
Random interaction network

9

University of Bologna

Ozalp Babaoglu

Cycle 180: Small Defective Clusters

10

University of Bologna

Ozalp Babaoglu

Cycle 220: Cooperation Emerges

11

University of Bologna

Ozalp Babaoglu

Cycle 230: Cooperating Cluster Starts to Break Apart

12

SLIDE 4

University of Bologna

Ozalp Babaoglu

Cycle 300: Defective Nodes Isolated, Small Cooperative Clusters Formed

13

University of Bologna

Ozalp Babaoglu

Cooperation Trend

14

% of cooperating nodes

University of Bologna

Ozalp Babaoglu

P2P File Sharing Application

Simplified version of a model by Q. Sun & H. Garcia-

Molina 2004

Nodes control how much capacity devoted to

generating or answering queries based on = [0...1]

=1.0 selfish (only generates queries)
=0.0 altruist (only answers queries)
We take as node utility the number of hits
Mutation of strategy: change randomly
Flood fill query forwarding, TTL’s, etc.

15

University of Bologna

Ozalp Babaoglu

P2P File Sharing Application

10,000-node network

Selfishness decreases Average performance increases

16

SLIDE 5

University of Bologna

Ozalp Babaoglu

Broadcast Application

Given a network of nodes, broadcasting involves

communicating a piece of information from a single node to all other nodes

To do this efficiently involves:
Minimizing the number of messages sent (MC)
Maximizing the percentage of nodes that receive the

message (NR)

Minimizing the elapsed time (TR)

17

University of Bologna

Ozalp Babaoglu

Broadcast Application

Given a network with N nodes and L links
A spanning tree has MC = N
A flood-fill algorithm has MC = L
For fixed networks containing reliable nodes, it is

possible to use an initial flood-fill to build a spanning tree from any node

Practical if broadcasting initiated by a few nodes only
In P2P applications this is not practical due to network

dynamicity and the fact that all nodes may need to broadcast

18

University of Bologna

Ozalp Babaoglu

Peer to Peer Networks

Peer to Peer (P2P) networks are generally:
Highly dynamic
Completely decentralised
Open
Untrusted
Hence many P2P networks generally settle for the

basic flood-fill approach or some variant of it

Superpeers topologies can improve the situation but

introduce network structuring overhead and require superpeers to act altruistically

19

University of Bologna

Ozalp Babaoglu

Overview of Results

Nodes have no incentive to forward a message once

they have received it

Doing so assumes altruistic behavior
Simulation results of broadcasting in unstructured and

dynamic networks where nodes behave selfishly

Even without altruistic behavior, broadcasting can be

sustained with L > MC > N

Hence, better than flood-fill, worse than spanning tree
However, on average NR = 85%
Hence without altruism, NR is far from optimal (many

nodes missing messages)

20

SLIDE 6

University of Bologna

Ozalp Babaoglu

The Broadcast Scenario

Nodes selected randomly to initiate broadcast (one every 10 cycles)
Node initiates a broadcast by sending a message to each neighbor
When receiving a message for the first time, node applies one of

two behaviors (strategies):

Forward the message to all neighbors (pass)
Do nothing (drop)
Utilities are updated as follows:
Nodes that receive the message gain a benefit
Nodes that pass the message incur a cost
Assume > > 0, indicating nodes have an incentive to receive

messages but also an incentive to not forward them

We used = 1, = 0.2

21

University of Bologna

Ozalp Babaoglu

Results

Initial network topology random
Each node initially set to pass behavior
Utilities initialized to zero
Considered two families of networks
static with no churn (as a baseline)
dynamic with high churn (1% of nodes leave and re-

enter each cycle)

New nodes link to randomly selected nodes from the

population and are randomly assigned a behavior {pass, drop}

22

University of Bologna

Ozalp Babaoglu

1000-node static random network

23

University of Bologna

Ozalp Babaoglu

1000-node high churn network

24

SLIDE 7

University of Bologna

Ozalp Babaoglu

Fixed random network

Average over 500 broadcasts x 10 runs

25