Thesis Proposal Data Consistency in DHTs

Background Peer-to-peer systems have become increasingly popular Lots of P2P applications around us –File sharing, file storage, Real time broadcasting, … Advantages of P2P system –Scalable –load balance, –fault-tolerant, –low-cost deployment Why is it so magic?

Core of P2P Each peer is a server and a client DHT –Every peer can find each other and cooperate –A hashed ID maps to a peer or an object –The object would be stored on the closest peer

Usages of DHT Directory service –Very little amount of information, such as URI, metadata, … –Don’t need to modified Storage –Data, such as files, … –Immutable, only for downloading and uploading What is the next?

Mutable DHT Most of DHT are for immutable data Immutable –Data wouldn’t be modified –Operations of DHT ping, search, publish Mutable –Data would be modified –Operations of DHT ping, search, publish, delete, modified

What is the challenge? Churn –Peer join, leave, fail randomly Replica –Fault-tolerant Problem –Object consistency Replicas are consistent on different nodes

What is the challenge? Network dynamic –Partial unreachability Problem –Key consistency Only one node is the root of a key

Related work Consensus system –Primary data maintainer –Clients modify the data on primary one –Do consensus algorithm After data modified After churn happened Quorum system –No primary one –Clients try to lock every replica before modifying

Related work Authorization –Contact the original closest one –Get Authorized token of the closer partition –Be the root of the partition

Questions Is key consistency really a big assumption on realistic environment? For what kind of churn and mutability, do we need a strong consistency mechanism? The importance of key and data consistency mechanisms –If we can guarantee only one of them, than what is the result on different situations?

Experimental Plan Create a realistic network environment for simulation Result of data consistency on different churn and mutability –DHT without any mechanism –DHT with key consistency mechanism –DHT with data consistency mechanism –DHT with key and data consistency mechanisms

Network Environment for Simulation Partial unreachability –Path outage probability of internet –Path outage duration and inter-duration Distribution of Node up and down –Statistics of real p2p network Topology –Hierarchical topology (few depth of AS’) –BRITE

References 1.Non-Transitive Connectivity and DHTs 2.Key consistency in DHTs 3.Toward Fault-tolerant Atomic Data Access in Mutable Distributed Hash Tables 4.Etna: a Fault-tolerant Algorithm for Atomic Mutable DHT Data 5.PaxonDHT: Achieving Consensus in Distributed Hash Tables 6.MUREX: A Mutable Replica Control Scheme for Structured Peer- to-Peer Storage Systems 7.A Practical Distributed Mutual Exclusion Protocol in Dynamic Peer-to-Peer Systems 8.An Experimental Study of the Skype Peer-to-Peer VoIP System 9.Actively Monitoring Peers in KAD 10.An Analysis of Network Reachability Using BGP Data 11.PlanetSeeer: Internet Path Failure Monitoring and Characterization in Wide-Area services 12.Quantifying the Effects of Routing dynamics on End-to-End Internet Path Failures

Next!! Questions –Nothing –Key consistency –Data consistency –Key and data consistency Narrow down –What my application like? –The parameters would be? –Why can’t Others experiment and parameters I use? –Why they are not realistic –Simulate the same other and change parameters to mine –Churn Distributions parameters –Experimental plans!! (all details)