1. Cs286r Victor Chan Scrip Systems Victor Chan

2. CS286 Victor Chan Agenda  Scrip Systems  Peer to Peer Systems  Scrip Systems for P2P Networks  Adobe Interna l 2

3. CS286 Victor Chan What is a scrip? A scrip is a non governmental currency used to pay for services from others. The need to earn scrip prevents freeloading  Adobe Interna l 3

4. CS286 Victor Chan Where is scrip systems used?  Capitol Hill Baby Sitting Co-op (Sweeney ‘77)  Couples babysitting for each other  Paid in scrips, each worth one hour of babysitting time  Low circulation of scrips resulted in “recession”  Eventually too much scrip was issued  Ithaca Hours (started in 1991)  Local currency used at Ithaca New York  500 business participating, including libraries, banks, medical centers, landlords  Used to promote local economic development, with12,000 Hours in circulation  Adobe Interna l 4

5. CS286 Victor Chan Yootopia! (Reeves et al 2006)  Yootle, a local currency created at Yahoo!  Used in prediction markets  Used to buy favors from people  Used where cold hard cash isn’t the best idea  All transactions are recorded in a ledger system  Group decision with a scrip system (Where to go for dinner?)  Voting with compensation  Vickery-Clarke-Groves (VCG) Mechanism  General Decision Auction (DAUC)  Iterative Decision Auction  SMS and web interface for users

6. CS286 Victor Chan Moving on  Any further thoughts on group decision or yootopia?

7. CS286 Victor Chan Peer to Peer Networks  Filesharing  BitTorrent, Kazaa, Gnutella, Napster  Online discussion  Slashdot, Digg, etc.  Distributed computing  Seti@home, Einstein@home

8. CS286 Victor Chan Peer to Peer Networks in Files Sharing  Increased social welfare  Costs still exist, leading to free riding users  Gnutella 70% users do not share, and 50% requests filled by top 1% users  There exist “altruistic” users that have become vital to the “health” p2p systems  However these users are expensive to host on a network and ISP’s are trying to remove them  Fair sharing does not happen since these users exist

9. CS286 Victor Chan Barter like approaches:  BitTorrent  Tit for Tat algorithm (Optimistic Unchoking)  Exchange upload bandwidth for download bandwidth  New peers lose out, nothing to offer  eMule  Track history of previous interactions with other users  Give priority to users with good history  With large n, hard to match up

10. CS286 Victor Chan Reputation based approaches:  Adobe Interna l 10  Internet Relay Chat (IRC)  Direct client to client sharing  Set up using private messaging/negotiation  Slow for new users to gain enough “rep”  Kazaa  Measure ratios of upload vs. download  To help new users, everyone is given “avg” rating  Free ride until “bad” rating, and create a new account (sybil attack)

11. CS286 Victor Chan Scrip system for P2P  Benefits of having a Scrip System  In history based reputation systems, no longer need to meet same peers  In BitTorrent, tit for tat can be extended to an exchange between multiple users  “The Role of Prices in Peer-Assisted Content Distribution” Johari et al  New users can be given scrip right away to participate  Problems of having a Scrip System  Still vulnerable to sybil attacks  How much money to have in the system?  Inflation, bubbles, recessions just like the real economy!

12. CS286 Victor Chan Scrip System in P2P networks  Efficiency and Nash Equilibria in a Scrip System for P2P Networks  Friedman, Halpern and Kash (2006)  Model for evaluating Scrip Systems in P2P Networks  Nash Equilibrium with threshold strategies  Money supply to maximize efficiency

13. CS286 Victor Chan The Model  Model Asymmetric interactions in a file sharing network  Unlike previous models of random matching between users  Each round uniformly select an agent to request and match with provider  Providers in the system each with β > 0 probability of fulfilling a request  Assumption: Time independence  Agent fulfilling request will pay cost α < 1  A discount factor of δ < 1 is used  Time steps are in 1/n

14. CS286 Victor Chan More definitions  G(n,δ,α,β) represents a game with n agents  : agent chosen in round t to make request  : whether a given agent can satisfy request, dependent on β  : whether a given agent will satisfy the request  : the agent chosen to satisfy request.  Chosen at random from willing and able  : agent i’s utility at round t

15. CS286 Victor Chan Utility functions  Standard User:  Altruistic User:  Total utility:

16. CS286 Victor Chan Altruistic Users: Closer look  Always happy to upload, since cost is positive and their strategy is  for all t  If enough of these users, others become free riders and play  for all t  How many altruists do you need to make everyone a freeloader?  Proposition 2.1: There is an a that depends only on δ,α,β such that in G(n, δ,α,β) with at least a altruistic users not volunteering is the dominant strategy for all standard users.

17. CS286 Victor Chan How many altruists?  With no money, users have their requests filled with probability:  So even with money, their total additional utility gain is:  But if this gain is less than the cost to get money: Users will not want to pay the cost and will never choose to volunteer

18. CS286 Victor Chan What does this mean?  Example with β = 0.01, α = 0.1, δ = 0.9999/day then need a>1145  Relatively small size compared to a large P2P network  In BitTorrent having 1145 Seeds (altruist) is unlikely, so we still see many leechers uploading.  Any thoughts on why amount doesn’t depend on n?  In order to establish a useful scrip system, need to remove altruistic users, or standard users will all become free riders

19. CS286 Victor Chan Finding the equilibrium in a Scrip System  Users pay those that satisfy their requests $1  Total amount of money in system M  Agents using threshold strategies:

20. CS286 Victor Chan Nonstrategic play of the game  System “converges” to a distribution over money  Assume everyone plays and system has M< kn dollars  State of the game can be represented as: Total amount of money in state s Player has value in this set

21. CS286 Victor Chan Distributions of Money in the System  Let be the distribution on {0,….,k}  Not very useful by itself, since not all distributions can be achieved  Look at the distributions that has, where m = M/n  There is a unique distribution in d*, with maximum entropy  Markov Chain,, then with large n, will likely be in a state s, such that d s will be close to d*  Closeness is defined as the Euclidean distance between two distributions: (

22. CS286 Victor Chan Theorem  X is the random variable that the Markov Chain is in a state S at time t  After some time t, the Pr(X is in state S where d s will be close to d*) is very high.

23. CS286 Victor Chan Simulation Results

24. CS286 Victor Chan Simulation Results

25. CS286 Victor Chan Simulation Results

26. CS286 Victor Chan Game under strategic play  Goal: Show that there is a non trivial Nash Equilibrium where all agents play a threshold strategy  First show for all k, if all other agents play S k there is a S k’ for agent i that is also the best response.

27. CS286 Victor Chan Make the strategies continuous  Look at a strategy pair, and consider a mix strategy  will play with probability and with  This essentially produces as continuous set of strategies by mixing adjacent threshold strategies. where

28. CS286 Victor Chan Theorem 4.1  If every other agent is playing then the best response is either a unique or a mix of playing two adjacent threshold strategies.

29. CS286 Victor Chan Proof of 4.1  Consider agent i with probability of making a request and receiving a request constant  i decides at each iteration whether to satisfy a request based on its strategy  So to i, the system is a Markov Decision Process, with i having the choice to move between various states  i will compute the optimal policy for this MDP, and there is a optimal policy that is a threshold policy.

30. CS286 Victor Chan Theorem 4.2  There should be a Nash Equilibrium that is in the space of threshold strategies  Fixing δ, we get a best response function that is a step function.  Any point where the br( δ, γ ) = γ then there is a Nash Equilibrium

31. CS286 Victor Chan Simulation Results

32. CS286 Victor Chan Social Welfare and Scalability  How much money M should be in the system?  Theorem 5.1: Most efficient equilibrium only depends on the ratio of M to n.  Proof, from Theorem 3.1, the d* depends only on M/n and k, and since br( δ, k ) depends on only d*, the Nash Equilibrium is only depend on M/n  In practice, it will be easier to adjust the price of a transaction rather than injecting or removing money from the system.  New comers can be added by changing the price of transactions

33. CS286 Victor Chan Sybils and Collusion  Sybils can be used to increase the likelihood of being chosen to fulfill a request  Set a lower k threshold strategy, offer to work more often  Sybils can also be used to drive down the price of requests  Or make sybils leave and drive up the price  Price of fulfilling a request depends on n

34. CS286 Victor Chan Extensions  The current system is Homogenous, relax these assumptions  Cost of joining the network, could deter sybil attacks  The current system does not take into account of altruistic users  Effect of hoarders, people who work but never spend (stocking up)  Any scrip system will require a centralized accounting system, and users will likely have to reveal their identities

35. CS286 Victor Chan That’s it! Q & A  Adobe Interna l 35