Irregular Graph Algorithms on Parallel Processing Systems.

Research Interests and Motivations My thesis work focuses on designing new parallel approaches for analyzing large real-world networks. This area of research specialization lies at the interface of high performance computing and big data analytics. What makes network analysis so interesting is the variety of data that can be expressed using graph abstractions. A graph representation fits naturally for biological networks at various scales, intercommunication in social networks, visual and geospatial relationships, neural activity in the brain, and numerous other sources of data within the scientific realm. The computational difficulties associated with analyzing real-world graphs is widely recognized, and it is correspondingly listed as one of DARPA’s 23 toughest mathematical challenges [1]. The high complexity, scale, and variation of graph-structured data poses an immense challenge in the design of techniques to study and derive insight from such data. Therefore, as we build more powerful supercomputers, it is important to understand how we can efficiently solve graph-theoretic problems on modern hardware. A 50× performance gap is sometimes observed between estimated and real running times due to programming model overhead, poor abstraction and subroutine choices, and sub-optimal data layouts. HPC platforms with manycore accelerators, such as GPUs or the Intel Xeon Phi, pose an entirely different set of challenges for algorithm designers to overcome. My primary research goals focus on tackling these problems through the optimization of graph analytics at all levels of hardware architecture, from thread to core to processor to single-node to multi-node to system-level scale. The fact that graph-structured data is so universal means that this research is useful to a large collection of data-intensive problems within the social and physical sciences. This abstract will describe some of the graph problems I’ve worked on throughout my doctoral program, their implementation, and their performance results relative to the prior state-of-the-art.