68. DynoGraph: Benchmarking Dynamic Graph Analytics
Authors: Eric Hein (Georgia Institute of Technology)Tom Conte (Georgia Institute of Technology)
Abstract: Large scale graph processing is the key to understanding complex relationships, ranging from the interaction of people on social media to the flow of data through a corporate computer network. Graph analytics present unique challenges for HPC system designers since they lack data locality and are difficult to partition into equally-sized units of work. In addressing these challenges, many researchers have opted to work with static graphs instead of the more difficult case of dynamic graphs that change rapidly during the analysis. Dynamic graphs require an entirely different memory layout, leading to degraded performance as algorithms traverse a fragmented, unsorted graph data structure. DynoGraph provides a standard for benchmarking dynamic graph analytics engines, bringing needed focus to this important class of applications.
Two-page extended abstract: pdf