Using Workload Characterization to Guide High-Performance Graph Processing

Mohamed Hassan, Virginia Tech
Endogenous contact networks generated by Argonne's CityCOVID model

Image: Argonne National Laboratory

Abstract: Graph analytics represent an important application domain widely used in many fields such as web graphs, social networks, and Bayesian networks. In this Big Data era, graph sizes increase explosively making it harder to efficiently mine and analyze these huge data-sets. Furthermore, the heterogeneity of the accelerators nowadays makes it more challenging to develop scalable graph processing frameworks that make efficient use of such increasing computational power. Hence, the ever-increasing graph sizes, and complexity of analytics algorithms brought this application domain to the forefront of high-performance computing. Following this, enhancing graph processing utilizing emerging accelerators and heterogeneous systems will be discussed. This talk explores FPGAs, CPU-FPGA hybrid systems, and quantum computing. The scope of this research extends to both, the performance and programmability (productivity) aspects of heterogeneous systems.

Please use this link to attend the virtual seminar: