Graph Analytic Hardware Acceleration via an OpenCL Shared Virtual Memory API

This research effort will develop an OpenCL Shared Virtual Memory (SVM) hardware acceleration Application Programming Interface (API) capable of enabling existing graph analytic APIs (e.g. GraphLab, GraphCT) to leverage a scalable, GPU-based hardware platform to significantly improve the performance of graph analytics including graph traversal and Breadth First Search (BFS). The use of OpenCL provides a hardware agnostic abstraction layer capable of exploiting heterogeneous, high-performance computing platforms that include flexible configurations of CPU, GPU and FPGA resources. OpenCL is a simple, common and open programming model that enables functionally equivalent algorithms to be deployed on heterogeneous computing architectures. Deployment decisions can be postponed until run-time allowing optimization of the deployment for the given scenario (e.g. high performance, portability, low-power). OpenCLs recently released SVM model, enables performance improvements for data-intensive and memory-bound applications like graph analytics, Therefore, heterogeneous hardware acceleration is now available for both computational intensive (i.e. compute bound), and data-intensive (i.e. memory bound) applications with the same common programming model (OpenCL). This research has immediate direct impact to current military situational awareness applications like cyber defense, and target identification and tracking, but also provides a solution for many commercial market needs including social network analysis (i.e. Facebook).