Title:  Graph Computing with Near-data Processing

Committee: 

Dr. H. Kim, Advisor     

Dr. Qureshi, Chair

Dr. Yalamanchili

Abstract: The objective of the proposed research is to provide an efficient graph computing framework with near-data processing. With the emergence of data science, graph computing is becoming an important tool for processing large-scale network data. Various graph computing frameworks have been proposed on both CPU and GPU architectures. However, because of the inherent irregular access pattern introduced by graph structures, graph computing suffers from significant inefficiencies in cache performance for CPU platforms, and memory divergence for GPU platforms. Meanwhile, reignited by recent advances in 3D-stacking technology, near-data processing (NDP) is getting more and more attentions. Prior works have demonstrated the potential of NDP offloading for improving performance of a number of graph workloads. However, it still remains an open question that how to realize a high performance graph computing framework with NDP. The needs of big data processing require immediate attention from architecture researchers to propose new software framework as well as new NDP architecture for efficient graph processing. The proposed research will perform a comprehensive study from both software and architecture perspective to enable an efficient graph framework with NDP techniques.