Physical Optics Corporation — Department of Defense SBIR Phase I: SB152-004

Description

To address the Defense Advanced Projects Research Agencys (DARPAs) need for next-generation computing systems for processing large scale graph datasets, Physical Optics Corporation (POC) proposes to develop a new Domain-Aware Self-learning Hardware Accelerated Graph Analytics (DASHGRAPH) system. This proposed solution is based on a new design that utilizes machine learning algorithms in a reconfigurable manner to greatly accelerate the performance of existing graph analysis software packages. Specifically, an innovative new architecture that minimizes the memory access bottleneck will enable the DASHGRAPH to offer a performance boost for a variety of graph types, including network structured, poorly structured, and applications with multiple data sets. As a result, this technology offers real-time graph analytics and ease of integration, which directly address the DARPA requirements. In Phase I, POC will demonstrate the feasibility of DASHGRAPH with detailed modeling and simulations of the system. During Phase II, we will develop a DASHGRAPH prototype at a technology readiness level (TRL)-6 and demonstrate its ability to use an existing graph framework like TinkerPop in a highly accelerated manner for a DoD-relevant problem like NetFlow data analysis.