OMNIA’s paper Accepted to ASPLOS 2019
Professor Myeongjae Jeon’s paper titled “HyStream: Stream Analytics on High Bandwidth Hybrid Memory” has been accepted to ASPLOS 2019. This work was done in collaboration with researchers at Purdue University, University of Toronto, and Google.
This paper presents HyStream, a streaming analytics system designed for machines with hybrid memory that combines DRAM with HBM (High Bandwidth Memory). To make best use of HBM, HyStream introduces a data structure called KPA (Key Pointer Arrays), proposes the use of sequential grouping algorithms, and provides a means to dynamically balance the use of HBM and DRAM based on resource drainage in the system.
ASPLOS is the top conference for multi-disciplinary computer systems research that spans across computer architecture, compilers, and operating systems.