PeerAssisted Content Distribution with Prices

1
Peer-Assisted Content Distribution with Prices

• Christina Aperjis, Michael J. Freedman, Ramesh
  Johari
• ACM CoNEXT 08

Presented By Michael Ma
2
Background Motivation (1)

• BitTorrent
• To incentivize uploads
• Bilateral barter-based system
• Rate based Tit-for-tat
• Problems
• Resource allocation
• Incentive mechanisms
• E.g. discovery of stable peering relationships is
  slow

2/15
3
Background Motivation (2)

• Why with Prices?
• Peer-assisted content distribution
• Nature of content sharing Supply and demand
• Employ Price Theory in Economics
• Due to market failure
• Hard to find good reciprocation
• Solution in economics multilateral exchange
• Multilateral exchange System matches user demand
  to available supply at other peers in the network

4
Contribution

• System Design
• PACE (Price-Assisted Content Exchange)
• Exchange currency for desired content with price
• Encourage sharing
• System Features
• Price management one price per peer
• Incentive benefits (credit can be spent later)
• Enable ISP to incorporate

5
Bilateral vs. Multilateral (1)

• Bilateral
• Peer i uploads to j iff j uploads to i in return
• Exchange ratio
• rij rate sent from i to j
• Peer prefer higher ratio to increase download
  rate
• BitTorrent

6
Bilateral vs. Multilateral (2)

• Multilateral
• Let a peer accrue capital by uploading
• Downloading can spend capital on anything
• E.g. peer i ? j, i ? k
• Comparison

7
Pricing Schemes

• How to price?
• One price per peer? (PP)
• One price per file? (PF)
• One price per file per peer? (PFP)
• Reason
• All equivalent if transfers only constrained by
  peer upload capacity
• PF worse than PP if topology is non-trivial
• PP and PFP yield equivalent equilibria, while PP
  simpler

v
8
Proof (1)

• Constraints
• Given download rates
• Ensure peers only upload files they possess
• Meet the band width constraint
• Budget constraints
• Ensure capital spends for downloading not exceed
  uploading
• Equilibrium
• The rate allocation r and peer price p that
  solves optimization problem
• PP, PF, PFP gets same equilibrium with the same
  rate allocation when constraint by upload
  capacity

9
Proof (2)

• If non-trivial network topology
• Links may be congested link price Pl
• PP, PFP schemes better than PF

4 peers, 2 files 2-cluster network Bi upload
capacity
No equilibrium for PF if Pl 0 for market
d1d3d2d4 ? pf/pg1 ? d18, not feasible ? max
rate 3
10
Hierarchical Network Model
Access link shared by all nodes in
cluster Immediate upstream link at s
11
PACE system

• Client Mechanisms
• Sell client
• Declare files willing to upload
• Total upload capacity willing to commit
• Price increase if demand exceeds supply, vice
  versa
• Buy client
• Choose files interested in
• Savings rate (for future)

12
System Design

• Procedure
• Peer requests download rates from other peers
• Pays uploading peer and links
• Payments to links are rebated equally to peers
• Functionalities
• Resource discovery
• Network friendliness (preferences)
• User management (registration, bootstrapping)
• Transactions currency security

13
Simulation Analysis (1)

• Prices for local links are cheaper
• When chunks are disseminated to a few peers,
  local prices drop

14
Simulation Analysis (2)
15
Discussion

• Evaluation
• straightforward
• Some are intuitive
• Professionally written
• Good and efficient
• Only design of system theoretical
• No consideration of ISP-propagated network prices
• Ideas are simple and efficient but not the first
  one

15/15