In this project, we are studying parallel, distributed, and stream processing techniques for managing large-scale graphs. Such graphs are becoming increasingly common in application domains like social network analytics, web graph analysis, bioinformatics, etc. These graphs can be highly dynamic, very large, or both dynamic and large. We investigate system-level as well as analytic techniques for addressing resulting challenges.

In this project, we look at system level techniques to improve performance of BSP style (Pregel-like) graph processing infrastructures on shared memory processors (SMPs).

In this project, we look at incremental maintenance of important graph structures (such as k-core, k-truss, communities, etc.) in the presence of dynamic changes to the graph in a distributed setup (for very large graphs).

Interaction graphs are append-only temporal graphs where interactions between entities are continuously added. In this project we look at streaming algorithms to best package edges to disk blocks, so that disk based interaction graph DBs can perform query processing more efficiently w.r.t. disk I/O.

• Ahmet Erdem Sarıyüce, Buğra Gedik, Gabriela Jacques-Silva, Kun-Lung Wu, Ümit V. Çatalyürek. “Streaming Algorithms for k-core Decomposition”, Very Large Data Bases Conference (VLDB), 2013.
• Buğra Gedik, Rajesh Bordawekar. “Temporal Storage and Querying of Evolving Interaction Graphs”, submitted to IEEE TKDE.

• Erdem Sarıyüce, Ohio State University
• Gabriela Jacques da Silva, IBM T. J. Watson Research Center
• Kun-Lung Wu, IBM T. J. Watson Research Center
• Qiong Zou, IBM China Research Lab
• Rajesh Bordawekar, IBM T. J. Watson Research Center