1. efstath@aueb.gr A Self-Managed Scheme for Free Citywide Wi-Fi Elias C. Efstathiou and George C. Polyzos Mobile Multimedia Laboratory Department of Computer Science Athens University of Economics and Business WoWMoM-ACC, June 13, 2005

2. efstath@aueb.gr Outline  P2PWNC Overview and Motivation  Background (hotspot market, P2P, and incentives in P2P)  P2PWNC Design Principles  P2PWNC Design  The NWAY Decision Algorithm  Simulations  Protocol and Implementation  Summary and Conclusions

3. efstath@aueb.gr The Peer-to-Peer Wireless Network Confederation (P2PWNC) 1 of 25 Manhattan WLANs, 2002 Skyhook Wireless Wi-Fi Positioning System (WPS) A wireless LAN (WLAN) aggregation scheme  Unites WLANs in citywide [con]federations  Requires no authorities  Relies on reciprocity between peers Motivation  Numerous WLANs, connected to the Internet, are within the range of passersby

4. efstath@aueb.gr Nokia 9500 Motorola CN620 Motivation (III)  WLAN-enabled mobile phones are on the market Motivation (IV)  Public WLAN operators mainly target “hotspots”  Municipal wireless still in its infancy Motivation (II)  Many WLANs are secured against outsiders  Need incentives to keep them open P2PWNC Motivation 2 of 25

5. efstath@aueb.gr P2PWNC: An incentives-based P2P system  Teams provide WLAN access to each other  Teams should provide in order to consume WLAN view Team view : WLAN access point : team member White team Green team Blue team The Rules of P2PWNC 3 of 25

6. efstath@aueb.gr Background

7. efstath@aueb.gr From Gartner:  2001: 1200 public hotspots worldwide  2003: 71 000 public hotspots worldwide  2005: 23 500 WLANs in hotels worldwide The Public Hotspot Market A subscription buys you (June 2005):  Sprint PCS: 19 000 hotspots worldwide  Boingo Wireless: 17 400 hotspots worldwide  T-Mobile HotSpot: 16 663 hotspots worldwide Skyhook Wireless data (2005):  50 000 WLANs in just 5 Massachusetts cities and towns (Watertown, Brookline, Roxbury, Newton, and Cambridge) 4 of 25

8. efstath@aueb.gr General term  Usually associated with file sharing systems  Also includes: Grids (computation) (Mobile) ad hoc networks (packet forwarding) Distributed Hash Tables (scalable, fault-tolerant storage) eBay-like communities (electronically mediated communities of providers and consumers) P2P Systems Distinctive characteristics  Peers act as both providers and consumers of resources  System relies on peer cooperation  Free-riding will prevail if: there is a cost involved with providing resources there are no authorities that can punish or reward exclusion from consuming the shared resources is impossible 5 of 25

9. efstath@aueb.gr Micropayments  Digitally signed tokens used as payment  Requires online bank to check for double spending (and to issue the credits) Yang, Garcia-Molina, “PPay: Micropayments for P2P Systems,” ACM CCS’03 Incentive Schemes for P2P (and representative papers) 6 of 25 Multiple account holders  Other peers maintain a peer’s account balance  Use majority rule in case of disagreement Visnumurthy, Chandrakumar, Sirer, “Karma: A Secure Economic Framework for P2P Resource Sharing,” p2pecon’03 Tamperproof modules  Each peer maintains its own account balance  Increase when providing, decrease when consuming Buttyan, Hubaux, “Stimulating Cooperation in Self-Organizing MANETs,” ACM/Kluwer MONET 2003

10. efstath@aueb.gr P2PWNC Design Principles

11. efstath@aueb.gr Why P2P?  A lot of underexploited WLANs out there set up by individuals  Hotspot operators (the corresponding “centralized model”): operate only a small fraction of the WLANs out there further segregate WLANs by competing for venues among themselves P2PWNC Design Principles 7 of 25 Micropayments, tamperproof modules, multiple account holders: Why choose another incentive scheme?  Require central authority (micropayments)  Are unrealistic (tamperproof modules)  Assume peers want to maintain accounts for others and/or perform auditing by trying to encourage “account holding” we get back where we started We need a simple incentive scheme that will encourage participation and cooperation even at the expense of accurate accounting

12. efstath@aueb.gr Adopt N-way exchanges as the incentive technique  A generalization of barter, which retains some of its simplicity  “Provide to those [who provided to those]* who provided to me”  A type of indirect reciprocity (sociology term)  Scales to larger populations, compared to direct-only exchanges  Does not require (central or distributed) authorities N-way Exchanges ABCD 8 of 25 Some variants of the basic N-way scheme: Cox, Noble, “Samsara: Honor Among Thieves in P2P Storage,” SOSP’03 Ngan, Wallach, Druschel, “Enforcing Fair Sharing of P2P Resources, “ IPTPS’03 Anagnostakis, Greenwald, “Exchange-based Incentive Mechanisms for P2P File Sharing,” ICDCS’04 Feldman, Lai, Stoica, Chuang, “Robust Incentive Techniques for P2P Networks,” ACM EC’04

13. efstath@aueb.gr P2PWNC Design

14. efstath@aueb.gr System Entities Team = Members + Access Points (APs)  Teams := P2PWNC peers  Assume intra-team trust  Team ID = (unique) PK-SK pair Member certificate  Member ID = (unique) PK-SK pair  Member certificate binds Member PK to Team PK Receipt  Encodes P2PWNC transactions between teams  Signed by consuming member  Receipt weight: amount of bytes the AP forwarded Member PK Team PK Member cert Timestamp Team PK Signed by Team SK Signed by Member SK Weight PK: public key SK: private key 9 of 25

15. efstath@aueb.gr Receipt Generation 10 of 25 C P CONN CACK 11:50am = t 0 (member connects) C P RREQ RCPT 11:51am (P requests 1 st receipt) RCPT timestamp = t 0 RCPT weight = w 1 C P RREQ RCPT 11:52am (P requests 2 nd receipt) RCPT timestamp = t 0 RCPT weight = w 2 > w 1 P RREQ RCPT 11:53am (member has departed) P stores last receipt (timeout) Receipt Repository

16. efstath@aueb.gr The Receipt Graph 11 of 25 A B C G H F E D I Directed weighted graph (with cycles) Vertices: team public keys Edge weight: sum of weights of corresponding receipts Edges point from the consuming team to the providing team W1 W2 W3 W4 W5W6 W7 W8 W9 W10 W11 W12 W13 W14 Graph security Free-riders and colluders can create an arbitrary number of fake vertices and edges They cannot create fake outgoing edges starting from teams who are outside the colluding group (they do not have the relevant private keys)

17. efstath@aueb.gr Receipt Repository 12 of 25 Two options:  Centralized repository Requires a well-known server that all teams can agree on All receipts are visible by all teams Server drops oldest receipts when full Mostly used to gauge the effectiveness of decentralized repositories Could have some practical importance  Decentralized repository Each team maintains its own private repository Fills it with receipts it receives during a WLAN transaction And with receipts it receives when gossiping Gossiping algorithm:  Roaming members carry receipts from their team repositories  They present them to the teams they visit  With RSA-1024 keys, a receipt is about 650 bytes long  With ECC-160 keys, a receipt is about 150 bytes long

18. efstath@aueb.gr Cooperation Strategies 13 of 25 Three cooperation strategies tested so far, each one:  Uses a different decision algorithm Input: the receipt graph Output: a decision of whether to provide service or not  May use a different gossiping algorithm (in the decentralized case) Different ways to choose the receipts that roaming members present  May use a different bootstrapping algorithm New teams need to provide before starting to consume For how long, and to whom? Specific decision algorithms include:  NWAY (assumes unit weights on receipts)  maxflow (borrowed from Feldman, Lai, Stoica, Chuang, “Robust Incentive Techniques for P2P Networks,” ACM EC’04)  gmf (generalized maximum flow) Progressively more robust against double-spending and collusion

19. efstath@aueb.gr The NWAY Decision Algorithm Searches for potential N-way exchanges  Red provides to Blue if there is a chain of receipts connecting Red to Blue  Red then discards all receipts in the discovered chain Team PK Member cert Signed by Member SK R: “B?” B G R Y X Z Team PK Member cert Signed by Member SK 14 of 25 Timestamp Weight Timestamp Weight

20. efstath@aueb.gr NWAY: Space Requirements Each team maintains 4 receipt repositories  IR – Incoming Receipts  OR – Outgoing Receipts  RR – Random Receipts  DR – Discarded Receipts holding up to s IR, s OR, s RR, s DR entries replacement rule: delete oldest receipt Each team has a Time Horizon (TH)  When DR overflows, TH holds the timestamp of the receipt that was just evicted  TH and DR allow ignoring all discarded receipts (at a cost…) time TH OLD TH NEW discardings evictions DR (filled to capacity) R’s repositories Hashes of discarded DR IRORRR R R 15 of 25

21. efstath@aueb.gr NWAY Operation Step 1. Provider searches for chain in merged repositories Temporarily merge IR C, RR C, RR P, OR P Consumer should carry IR C, RR C : space requirements? Consumer information revelation: incentives? Step 2. Provider discards receipts and updates TH First receipt can simply be deleted from OR P Provider may still want to send locally discarded receipts to its own roaming members Step 3. Provider and consumer store new receipt Consumer sends it to home OR C : incentives? Step 4. Receipt dissemination (as a side-effect) Provider updates RR P with receipts from IR C and RR C PC PC IR C RR C +RR P OR P …… … … … … … …… …… PC is visiting 16 of 25

22. efstath@aueb.gr Evaluation Framework Used to evaluate the performance of the 3 cooperation strategies against each other and against 3 strategies that do not rely on feedback (ALLC, ALLD, RAND)  Teams are randomly paired 1 round : every team gets one chance to consume, one to provide if a team decides to provide service, it loses c = 1 points if a team is provided service by another team, it gains b = 7 points  Peers may learn – evolve towards the highest-rated strategy from the ones available With probability proportional to the difference in rating  A strategy’s rating := the average of the running average score/round of its followers weighted according to how many rounds they have been using the strategy We also simulate system growth  Start with two teams; a new team joins at the end of each round; teams never leave NWAY parameters for the experiments that follow: s IR = s RR = s DR = 100, s OR = 400 17 of 25

23. efstath@aueb.gr NWAY against ALLC, ALLD 18 of 25

24. efstath@aueb.gr NWAY against ALLC, ALLD, RAND 19 of 25

25. efstath@aueb.gr NWAY against Traitors 20 of 25

26. efstath@aueb.gr Design Summary 21 of 25

27. efstath@aueb.gr P2PWNC Implementation

28. efstath@aueb.gr P2PWNC Protocol Messages (1/2) 22 of 25 CONN  Sent by the roaming member to the AP  Contains the member certificate (base64-encoded) CACK  Positive or negative response to connection request, sent by the AP  If positive, contains timestamp of session and public key of the providing team RREQ  Request to sign a new receipt for this session, sent by the AP  Contains the volume of traffic, as measured by the AP

29. efstath@aueb.gr P2PWNC Protocol Messages (2/2) 23 of 25 RCPT  Receipt (base64-encoded)  Sent by members to APs and by APs to the repository UPDT (only in distributed repository mode) Request by member to home repository for an update with the latest receipts (newer than a timestamp) QUER and QRES (algorithm specific) Communication between AP and home (or global) repository. The request contains the public keys of the (prospective) consuming and providing teams. The response contains a PROVIDE or DO_NOT_PROVIDE reply

30. efstath@aueb.gr Linux-based WLAN access point  We implemented the P2PWNC protocol (AP side) on it  32 MB RAM, 8 MB Flash, 200 MHz CPU  Retails for less than $70  Cryptographic, maxflow performance comparable to 200 MHz PC  Can act as home repository (storing more than 10 000 receipts) Linksys WRT54GS 24 of 25

31. efstath@aueb.gr Conclusion

32. efstath@aueb.gr Final Points and Summary 25 of 25  Hurdle to P2PWNC deployment: ISP sharing prohibitions?  Centralized or decentralized deployments?  People living in the outskirts: will the notion of teams be enough to incorporate them? (In all 3 decision algorithms, teams end up having to provide approximately as much as they consume – how will this work within a team?)  We demonstrated a family of practical incentive techniques for WLAN sharing  Teams do not have to trust one another!  There are no hard service guarantees  More at http://mm.aueb.gr/research/p2pwnc/

33. efstath@aueb.gr Thanks! Elias C. Efstathiou Mobile Multimedia Laboratory Department of Computer Science Athens University of Economics and Business http://mm.aueb.gr/people/efstath/