Peertopeer and Wi Fi in Mobileman Cambridge 3703
Peer-to-peer and Wi. Fi in Mobileman – Cambridge 3/7/03 Jon Crowcroft The Computer Laboratory University of Cambridge jon. crowcroft@cl. cam. ac. uk 1
People and Projects in Cambridge • Faculty: • Jon Crowcroft, Ian Pratt • Research Staff: • Sven Ostring, + • Ph. Ds: • Rajiv Chakravorty & Meng How Lim • Daavid Spence, Andrew Twigg, Tim Moreton • Projects: • • COM(vodafone) CMI Mobile. MAN Intel (marcello Pias_ 2
Relevant p 2 p technologies • Pastry/Scribe/Xenostore – • Storage, notification, searching • Consistent Hash + location • +++ Its Local +++ • Eternity, Freenet • Privacy, persistence, caching • Robustness • Rabin fingerprint, bloom filter • Xen – OS which can run XP, Linux, BSD, Symbian, JVM…(c. f. vmware, denali) • Seek Input from partner on Middleware 3
Location Services Experiment • P 2 p makes use of techniques such as consistent hashing, rabin fingerprints, bloom filters and so on, for mapping key to location • Perhaps a similar model could be used for addressing • Would note that these often lead to very poor latency • Techniques such as landmarks, shortcuts, lighthouse, chord etc are worth looking at • Paper in to IPTPS on “lighthouses” – MSR now implementing in Pastry (to get proximity closer to “triangle” law… 4
Notification Services • Could look at Scribe (P 2 P based Event service based on Pastry) • Could look at event services as driving model – could be information or applications like games…. • enlist Ph. Ds at cambridge!!!… • Depends on input from SUSPI on application of intererest 5
Other Applications on the Move • Notification (context – e. g. nearest gas station, nearestaurant) • Streaming (p 2 p – see later) audio (prob not video for cars content share, or server based slipstream is a Pastry service that does this • Collaborative tools (field workers – engineer working on fixing problem with other engineers remotely helping) • Other? (characteristics, security, cost etc) e. g. see Pastry (Xenostore and Xenostrust and Xenosearch) • Seek input from partner on Applications! 6
3 layer architecture • 3 Layer of System: • Ad hoc wireless • P 2 p information layer • Application layer • Two problems to tackle • Align incentives across layers • Deal with heterogeneity 7
Align Incentives • Use Wi. Opt paper model for p 2 p layer as well as for ad hoc wireless • Control law for earning and spending tokens • Control law for discounting tokens over time • Control law for minting of new tokens as population grows • Tokens for sharing spectrum, storage, content etc 8
Heterogeneity • Not all people have same resources • Not all people are so honest • How to create a system where equilibrium is close to trust and even share, but policy can be used to weight share? • Relay on Wiopt algorithm property of approximate cconservation of token: • In any region, after any change of population, and after usage has gone quiet, the system will reach equilibiu => • Can police propert • Central agency (regulator) • Distributed (witnesses) 9
Tolerating cheats, supporting the poor • Strategy for treatment of cheats (deteected by regulator or witnresses consensus that more tokens in use than should be: • Ostravcisum (“tit-for-tat”) – one shot – not long term • Strategy for supporting the poor: • Subsidy/Tax – weight tokens (change rate of accrueing and discounting of tokens (similar to proposals for service differentiation using weighted ECN pricing in TCP!) 10
Heterogeneity in P 2 P • It has been observed that “free riders” are not 100% idle in the community – they often just reduce what they offer to a lower %age than a team player would • This may reflect poverty (dealt with before) or meanness (also dealt with) • Main extension to Wi. Opt ideas are that we may have to factor in more than 2 parameters (capacity, battery, storage, etc) 11
Ongoing work with MSR and Intel • See if can add token scheme to multihop version of MAC • See if can add token scheme to pastry • Use in one application – one of • Scribe (event architecture) • Splitstream (media streaming with layered adatpive Qo. S) • Again, need to consider User Community • Possible synergy with MMAPPS and Equator 12
Other Cambridge Efforts • Hosting Wi. Opt ’ 04 • Infocom ‘ 04 2 submissions on seamless handover from gprs and 802. 11 • Several (4) posters and papers (5) in SIGCOMM 13
- Slides: 13