Jiwon Seo : 2013 Mobile and Social Workshop


Thursday, April 18, 2013
Location: Fisher Conference Center, Arrillaga Alumni Center

"SociaLite: A Datalog-Based Language for Efficient Social Network Analysis"


With the rise of online social networks services, distributed graph analysis is becoming very important. The current approach to large-scale graph analysis employs vertex-centric programming model, which requires programmers to implement low-level message passing code. We designed and implemented SociaLite, a distributed query language for graph analysis. SociaLite allows programmers to implement analysis programs in high-level semantics, and compiles the high-level programs into optimized low-level code. For distributed programming, SociaLite programmers simply annotate how data is to be distributed; the compiler automatically infers the necessary communication and generates the communication code for distributed multi-core machines. Recursive queries are supported in SociaLite, which enables analyses requiring recursive node traversal, such as finding single-source shortest paths. In our experiment, compared to Giraph, an open-source graph analysis framework with vertex-centric programming model, SociaLite programs are an order of magnitude more succinct. Our performance evaluation shows that this high-level semantics of SociaLite does not incur any performance penalty.


Jiwon Seo is a PhD student working with Professor Monica Lam. His recent research topic is parallel query language called SociaLite, which is designed for large-scale graph analysis. He works on all aspects of SociaLite, including the design of the language and the implementation of the compiler/runtime system.