repeated game modeling of multicast overlays mike afergan (mit csail/akamai) rahul sami (university...
DESCRIPTION
Application-Layer Multicast Position in tree can impact QoS. [Mathay et al 04] Users have motive and means to alter tree. In the limit, becomes the unicast tree. …… Wants to move up tree Want fewer children Problem: Selfish users can degrade system performanceTRANSCRIPT
Repeated Game Modeling of Multicast Overlays
Mike Afergan (MIT CSAIL/Akamai) Rahul Sami (University of Michigan)
April 25, 2006
Talk Overview Introduction Repeated Games A Repeated Game Model of
Multicast Overlays Results Summary
Application-Layer Multicast
Position in tree can impact QoS. [Mathay et al 04] Users have motive and means to alter tree. In the limit, becomes the unicast tree.
……
Wants to move up tree
Want fewer children
Problem: Selfish users can degrade system performance
A Double ProblemSystem Design Problem
Goal: A protocol which creates efficient trees even with selfish users.
This problem is hard: Real-time and unidirectional Heavyweight solutions (e.g., payments, complicated
trees) are undesirable. NATs make many solutions (e.g., monitoring) challenging.
Modeling Problem On a small scale, these trees exist in practice without
such mechanisms. [Chu et al ’04]Goal: A model that explains observed behavior and
provides practical insight for building robust protocols.
Key Insight
Cheating degrades system efficiency and quality
Can reduce lifespan of system
Even selfish users want the system to exist in the future.
Key Contributions A repeated model of cooperation
Cooperation is endogenous to model Does not require heavyweight
mechanisms
Prescriptive results for building more efficient systems
Talk Overview Introduction Repeated Games A Repeated Game Model of
Multicast Overlays Results Summary
One-Shot Prisoner's Dilemma
P1\P2 C D
C (5,5) (0,9)
D (9,0) (1,1) Static EquilibriumOutcome
In the one-shot game, (D,D) is the outcome of the unique Nash Equilibrium.
Repeated Prisoner's Dilemma
P1\P2 C D
C (5,5) (0,9)
D (9,0) (1,1)
$$$ or $+$+$+ $+ $ + S
Key Takeaway: The equilibrium of the repeated game may differ from the equilibrium of the stage
game.
Example Strategy: 1. Play C 2. If the other player defects, play D forever
Outcome ofthe RepeatedGame
Sample AnalysisP1\P2 C D
C (5,5) (0,9)
D (9,0) (1,1)
$$$ or $+$+$+ $+ $ + S
Parameterized by discount factor () Patience Factor (infinite game) Probability of game ending (finite game with unknown horizon)
Example: is an equilibrium of the RPD iff: (Playing forever) (One-time “cheat”) + (Resulting payoffs)
“Play C forever. If other plays D, play D forever” is an equilibrium iff:
10
)1(95t
t
t
t ½
Talk Overview Introduction Repeated Games A Repeated Game Model of
Multicast Overlays Results Summary
Model Intuition Nodes in a network form an overlay. Per time-period benefit to user dependant on:
Quality of content received Load on user
Network Efficiency: Relative network load of given tree Defines per-period probability of network
continuing Selfish players maximize the (discounted)
series of per-period payoffs.
Formal Game Model Instance
Network: G = (V,E) Nodes to be served: N V Single source: s N, sV Single atomic piece of content
An algorithm constructs a tree (T) which serves all nodes, N.
Load of tree L(T, G) is sum of load on all links.
Players and Actions User Utility Function – ui(di,ci)
Decreasing in d and c as fixed and exogenous Action Space: {Connect to Root, Drop Child,
Stay} Response Function – R(L)
1. R(L(Faithful Tree)) = 1.02. 1.0 > R(L(Unicast Tree)) ≥ 03. R(L) is monotonic
Equilibrium Condition: ''
1
''
0
,)'(,,)( iiit
ttiii
tiii
tt cduLRcducduLR
Talk Overview Introduction Repeated Games A Repeated Game Model of
Multicast Overlays Results Summary
Simulator1. Take inputs (topology, ui(.), N, , A)2. Randomly select source and N end-nodes.3. Each node learns di , ci, and f(L).4. Each node can connect to root, drop child,
or take no action.5. Repeat Step #4 until stable.6. Collect Statistics. All datapoints represent 90 simulator runs.
We prove that stable points of simulator are sub-game perfect equilibria.
Results
1. System efficiency decreases with decreasing .
2. System efficiency decreases with increasing N.
3. Specific insight for particular tree formation protocols.
Goal: A model that explains observed behavior and provides practical insight for building robust protocols.
Benchmark Algorithm:Naïve Min Cost Spanning Tree Inputs:
Nodes Pair-Wise distances
Outputs: Min Cost Spanning Tree
Assumes all reports are truthful
NICE (Banerjee et al ’02)
Nodes create hierarchical tree of clusters of size k Completely distributed NICE has been shown to have good performance
characteristics. [Banerjee et al, ’02]
NICE is more efficient than a Naïve Min-Cost Spanning Tree
NMC better for faithful users
But for even mildly selfish usersNICE performs better.
Utility DistributionNaïve Min Cost NICE
NICE has an inherent tradeoff between depth and load.
050
100
150
200
250
300350
400
450
4 5 6 7 8 9
UtilityCount
050
100150
200
250300
350400
450
4 5 6 7 8 9
Utility
Count
1
1.5
2
2.5
3
3.5
4
10.9
80.9
60.9
40.9
2 0.9 0.88
0.86
NMC
32
8
2
Impact of Cluster Size
Under reasonable assumptions, increasing clustersize can increase efficiency.
Load
Generalizations Core results and intuition apply to
more general cases: Large class of utility functions Large class of response functions Noisy signal of state Noisy understanding of response
function
Exogenous Types vs Endogenous Motivations
Prior models use exogenous types: Cheater/not [Mathy et al] Altruism parameter [Feldman et al ‘04, Chu/Zhang ‘04 ]
A repeated game model captures these factors in an endogenous fashion.
Benefits: Fewer degrees of freedom Behavior is dependant on the system.
This enable practical conclusions.
Summary Users have the means and motive
to alter multicast overlay trees. A repeated model of interactions
can explain user cooperation without heavyweight mechanisms.
Behavior which is endogenous to the model enables practical conclusions.