Things to Consider to Enable Dynamic Graphs in Processing-in-Memory

Abstract

With the ubiquity of graphs to represent large, sparse data sets, a recent focus has been placed on making graph processing more efficient. Processing-in-Memory (PIM) is an alternative solution to reduce the data movement between memory and processors, and it results in better performance and reduced energy consumption. However, prior PIM-based graph processing work operated only on static graphs. In fact, real-world graphs are constantly changing as their data is updated. In this paper, we will discuss design considerations for adapting dynamic graph processing to PIM.