행사세미나 University of California 전상우 교수 세미나
페이지 정보
본문
일시:: 7/13(수) 오후 4시
장소: 화학관 330110
· Abstract:
Out-of-core analytics, where data of interest is stored and processed in secondary storage, can handle much larger data sizes than in-memory approaches due to the relative cost-effectiveness of storage. While modern high-performance solid-state storage devices can often deliver sufficient streaming bandwidth to keep computation units busy, it can still become the performance bottleneck for complex analytics applications which are either fine-grained random access intensive, due to the access granularity mismatch. In this talk, I present the design pattern of using hardware acceleration to re-organize fine-grained access patterns, and demonstrate its benefits on some prominent applications spanning graph analytics, bioinformatics, and databases, resulting in affordable out-of-core analytics with no performance loss compared to in-memory systems.
Sang-Woo Jun is an assistant professor in Computer Science at the University of California, Irvine. He obtained his Ph.D. from the Department of Electrical Engineering and Computer Science at MIT in 2018 on the topic of near-storage acceleration for Big Data analytics. His research focus is on efficient use of memory and storage in the context of application-specific hardware accelerators, and demonstrated orders of magnitude performance improvements compared to conventional approaches.