Dr. Esther Pacitti, LINA and INRIA, University of Nantes

Data replication and caching in P2P systems

February 27th, 2009
10.00 pm
Seminars room, LIRMM

Abstract:
Data replication and caching are two well-known, complementary approaches to improve data availability and access performance in distributed systems. In the context of P2P systems which are highly dynamic and large-scale, supporting data replication and caching efficiently is a hard problem. In this talk, I describe our recent results on this problem in two research projects: P2P-LTR and Flower-CDN.
P2P-LTR is a P2P Logging and Timestamping service for Reconciliation of replicated data in the context of P2P collaborative text editing. The challenge is to support concurrent updating at multiple peers and the dynamic behavior of peers that may join or leave the network at any time during concurrent editing. Our solution enables an existing reconciler (based on operational transforms) to operate in a P2P environment using continuous timestamping of update operations. It exploits a distributed hash table (DHT) to log update operations and assign timestamps to updates in a fully distributed and highy available way. In the context of the European STREP Grid4All project and the ANR Xwiki Concerto project, we validated P2P-LTR by an implementation on top of the Chord and Pastry DHTs for a P2P wiki application. Our performance evaluation shows the scalability of our solution.
Flower-CDN is a P2P content distribution network (CDN) which we are developping in the context of the ANR Dataring project which targets online communities such as social networks. It is aimed as a light-weight alternative solution to more specialized, heavy weight CDNs such as Akamai. Flower-CDN is a locality-aware P2P caching system in which the users that are interested in a website support the distribution of its content. The idea is that peers keep the content they retrieve and later serve it to other nodes that are close to them in locality. Flower-CDN exploits a hybrid architecture with a DHT and gossiping. When a node requests some content from a website for the first time, a locality-aware DHT quickly finds a peer in its neighborhood that has the content available. Additionally, all peers in a given region that maintain content of a particular website build an unstructured content overlay (called Petal). Within a content overlay, peers gossip information about their content allowing the system to maintain accurate information despite failures and churn. Compared to a popular CDN which DHT-based only, Flower-CDN reduces lookup latency by a factor of 9 and transfer distance by a factor of 2. Furthermore, Flower-CDN's gossip has low overhead and can be adjusted according to hit ratio requirements and bandwidth availability.