A graph is a ubiquitous data structure that models entities and their interactions through the collections of nodes and edges. It is widely employed in several important application domains ranging from social media, navigation tools, search engines, physics simulations, and biology. Despite its prevalence, the performance of graph workloads on commercial platforms is limited. This is mainly due to the irregular nature of memory accesses and convoluted control flow instructions used in graph applications while accessing large amounts of real-world graph data (i.e., billions of nodes/edges). Therefore, there is a pressing need for optimizing the performance of graph workloads.
In this talk, I will present a systematic optimization study of graph workloads running on both static and dynamic graphs. Specifically, I will present two of my most recent works called NDMiner [ISCA 2022] and Mint [MICRO 2022] in detail. NDMiner optimizes the execution of the Graph Pattern Mining (GPM) application. In this work, I will showcase how to combine the benefits of Near Data Processing (NDP) and domain specialization to improve GPM workload performance. Mint, on the other hand, investigates and optimizes a pattern mining application on temporal graphs (a type of dynamic graph), called temporal motif mining. Mint presents a new programming model, hardware accelerator architecture, and domain-specific optimization to significantly improve the performance of mining temporal motifs. In addition to these works, I will briefly tough upon our optimization and detailed benchmarking effort for traditional graph processing algorithms (e.g., PageRank and SSSP) and random walk-based graph learning pipelines (e.g., node classification and link prediction).
Nishil Talati recently completed his PhD from University of Michigan, USA. Prior to joining the PhD program, he completed a master’s degree with thesis from Technion, Israel, and an undergraduate degree from BITS Pilani, India. Nishil’s research interests span computer architecture, compilers, and software engineering. Specifically, he has worked on several optimization efforts to improve the memory performance of modern computing systems. One of his recent works was recognized as the best paper award at HPCA 2021.
Important: The participation is free of charge, but registration is required
For more details and updates on the series of “ACRC Semiconductor Webinars” please follow our newsletters and our website