Google Tech Talk (see below) June 17, 2010 by Josep M. Pujol. ABSTRACT The difficulty of separating the social graph has brought new challenges of designing systems for the scale of online social networks (OSN). The vertical scale for the use full replication can be an expensive proposition. Horizontal scalability by partitioning and distributing data across multiple servers using, for example key-value store with DHTs can suffer costly inter-server communication and provide differentTopics. These challenges are often led to costly re-architecting efforts to OSNs popular as Twitter and Facebook. We design, implement and evaluate SPAR, a socio-partitioning and replication middleware that mediates between application and database-level OSN. Spar used to partition the underlying social graph data structure, the user and selectively replicated to ensure users that their neighbors have user data on your machine co-located. The profits of these are multi-fold:Application developers can expect the local semantics, ie to develop, as for a single machine, scalability is achieved with the addition of commodity machines with low memory requirements and network I / O, and N + K redundancy reached at a fraction of the cost. We offer a complete design system, a comprehensive analysis of data from Twitter, Orkut and Facebook, and implementation work. We have shown that SPAR works well in terms of reducing overhead costs and dealing with high dynamic...
Keywords: google tech talk, social networks
No comments:
Post a Comment