Speaker
Description
In today's data-driven world and heterogeneous computing environments, efficient analysis of large-scale graphs has become more important than ever before. Therefore, there has been a significant interest in high-performance graph processing. On one side, many researchers developed optimized algorithms for specific graph analytics kernels for specific architectures. On the other side, software systems and graph databases have been designed to leverage modern high-performance computing platforms. Some of them provide a very productive programming environment for graph analysis, however, they cannot get even close to the single-threaded performance. I will highlight the key optimization ideas and patterns, on a few examples of hand-optimized kernels. As it will be evident, almost all of the optimizations revolve around data movement. Then briefly, first, I will present a new block-based graph algorithm framework that offers a sweet spot between efficient parallelism and architecture agnostic algorithm design for a variety of graph problems. Second, I will talk about challenges of deploying such graph analytics kernels in current graph databases.
Fiziksel Katılım:
Sabancı Üniversitesi, Altunizade Dijital Kampüs ROOM 202
Webinar ID: 927 6794 3429
Passcode: 489813
https://sabanciuniv.zoom.us/j/92767943429?pwd=SWNzRmt1S1NmQlRNSzZMVjdabWhJUT09
