Data/Storage Management The Group: John Kubiatowicz, Barbara Liskov, Scott Shenker Byung-Gong Chun, Alan Mislove, David Oppenheimer, Emil Sit, Hakim Weatherspoon
Applications The big thing about P2P: Management –Adaptation to failure –Self-Configuration Spectrum of Applications –Low Update, High Speed Reading –High Durability –Transactional applications Grand Challenges –Google-like high read performance –Internet Archive –Physics Applications –Amazon
What Semantics do we want from storage? Multiple Writers/ Multi-Object Transactions Performance –Caching Streaming –Effective event delivery –Effective update Delete Can we trust that data will still be there 10 years from now?
Open Question: Location control: –Is it important to place data? –Most “current apps” seem to desire control? –Most P2P solutions do not give control Implications of a layer of indirection –Is a layer of indirection important? –Is it fundamental?
Requirements for Real Storage Introduction Service –Who to speak to? –Are they the most current incarnation? Secure Routing: –Can we actually abstract away the routing Managed Infrastructure? –Much of our discussion seemed to assumed this –VS: Flakey Peer2Peer model
Desirable Services Management/diagnosis mechanisms Repair triggering facilities
Query the Data What types of queries should we support? –Efficient Joins vs –Data status and performance collection Reliability semantics –What happens under churn? Is this possibly universal? –Are there a set of querying mechanism that can be good for many apps?