PPT

Game Authority for
Robust and Scalable Distributed
Selfish-Computer Systems
Shlomi Dolev, BGU (Israel)
Elad M. Schiller, Chalmers (Sweden)
Paul G. Spirakis, CTI (Greece)
Philippas Tsigas, Chalmers (Sweden)
Distributed
Let’s play
computing
the prisoner
assumes
identical
dilemmaprograms.
game.
How to design distributed
algorithms for the wild internet?
Use game theory for selfishcomputer systems.
The system designer
Game theory predicts:
selfish-computer choose
betray!
John Nash
A
B
OK... Let’s playSilent
in a real system.
Silent
Betray
Yes, I trust
game theory.
Betray
The system designer
O.K., we should
enforce
In explicitly
a real system,
the implicitwhere
rules of
thethe
game.
I am
only
authority…
they would be free to
escape!
We need a distributed
game authority.
The system designer
The Society Moral Code
• Complete anarchy exists without moral codes
• Game authority founded over the moral majority
– choose and enforces the rules of the game
– promote freedom of choice for the society’s benefit
• We promote honestly selfish
behavior for the sake of:
• end-point creativity
• motivation for success
The Society Moral Code
• Complete anarchy exists without moral codes
• Game authority founded over the moral majority
– choose and enforces the rules of the game
– promote freedom of choice for the society’s benefit
Benefits
• End-point success
• that yields global success
• Provable scalability
• from the days of Greece
• Provable robustness, still
Technical Contributions
Cost Reduction:
• We replace the higher price of anarchy
with the lower price of stability!
Technical Contributions
Cost Reduction:
• We replace the higher price of anarchy
with the lower price of stability!
Price of anarchy (PoA)
Social optimum
Worst NE
• Worst case ratio between:
NE’s social cost, and
the social optimum
Koutsoupias & Papadimitriou STACS’99
PoA
Good
Bad
Technical Contributions
Cost Reduction:
• We replace the higher price of anarchy
with the lower price of stability!
Price of stability (PoS)
• Best case ratio between:
NE’s social cost, and
Social optimum
PoS
the social optimum
Anshelevich et al.
Best NE
Worst NE
PoA
FOCS'04
Good
Bad
Game Authority Implementation
• Can we assume that all components are selfish?
– impossible: Phy. layer game & Mac layer game &, … ,
& possible failures & imprecise utility
• how to bound the PoA?
• Honest and moral based
middleware tolerating
• Byzantine faults
• transient faults
• Facilitates interaction among
honestly selfish agents
Application-layer:
Social optimum
Best NE
PoS
Worst NE
∞
Honestly selfish agentsExplicit
(majority)
PoA
Middleware: Game Authority
Good Moral Code
Bad
Implementation (cont.)
• How to decide on the preferable game?
• How does the honest majority audit the game?
• How to preserve privacy in simultaneous plays?
• Byzantine agreement
• Cryptographic primitives
• Game theory analysis
Your attention is
appreciated
More details:
Technical report number TR-2006:9
Computer Science and Engineering
Chalmers University of technology, 2006
Also, technical report, DELIS, 2006. Accessible via
http://delis.upb.de/docs/
Rabbi Akiva said: All is foreseen, but freedom of
choice is given. The world is judged in
goodness, yet all is proportioned to one's work.
(Mishnah Pirkei Avot, Chapter 3, 19)
‫ והכול לפי‬.‫ והרשות נתונה; ובטוב העולם נידון‬,‫הכול צפוי‬
)‫ ג` ט"ו‬,‫ (פרקי אבות‬. ‫ אבל לא על פי המעשה‬,‫רוב המעשה‬