Designing Hybrid Architectures for Massive-Scale Graph Analysis

Turning large volumes of data into actionable knowledge is a top challenge in high performance computing. Our previous work in this area demonstrated algorithmic techniques for massively parallel graph analysis on multithreaded systems. This work led to the development of GraphCT, the first end-to-end graph analytics platform for the Cray XMT and x86-class systems with OpenMP, and STINGER, a high performance, multithreaded, dynamic graph data structure and algorithms. Both of these packages are freely available as open source software. This dissertation research culminates in experimental and analytical techniques to study the marriage of disk-based systems, such as Hadoop, with shared memory-based systems, such as the Cray XMT, for data-intensive applications. David Ediger is a fifth year PhD candidate in Electrical and Computer Engineering.