SPAR Scaling Online Social Networks
The sheer size of today’s online social networks is introducing a number of new system design challenges in scaling, management, and maintenance.
By “scaling” we mean the ability to provide a single month in 2009! The fundamental a good service to millions of users, while issue that one needs to deal with when dealing with immense popularity that scaling social networks is that by design could stress the system’s resources. It was those networks feature a number of reported that Twitter grew over 1300% in interconnections that cannot be easily partitioned into smaller groups, that could for instance allow one to host different parts of the graph on different servers. The aim would be to develop a system that is able to respond to a user’s question using the resources of a single machine, which in turn means that all required information to answer such a question is quickly accessible by that machine, in the best case local.
We have developed SPAR, a middleware solution that is able to transparently scale OSNs to hundreds of millions of users. SPAR stands for social network partitioning and replication. Instead of viewing the underlying interconnected data components to be a hindrance, we take advantage of the fact that such interconnections manifest in tight social communities. These tight social communities can be partitioned and by replicating nodes that lie in multiple communities, we can ensure locality of data, thus aiding in easier scaling.
We have studied SPAR using real datasets from three different OSNs - Twitter, Facebook and Orkut. SPAR provides locality semantics at the expense of moderate replication overhead.
Hence there is a tussle between online application providers wanting to exploit more personal information and end-users who want to protect that information...
- Vijay Erramilli, researcher.
- Georgios Siganos, researcher.
- Xiaoyuan Yang, researcher.
- J. M. Pujol, researcher.