Blogs (2) >>
ISMM 2017
Sun 18 Jun 2017 Barcelona, Spain
co-located with PLDI 2017
Sun 18 Jun 2017 15:00 - 15:30 at Aula Master - Session 3: Hybrid Memory Systems Chair(s): Ben L. Titzer

Modern applications running on supercomputers have increasing demands for memory, in terms of capacity, speed, power efficiency, and persistence. Since there is no single memory technology today that can satisfy all these requirements, next-generation supercomputers are expected to feature deeper memory hierarchies that consist of characteristically different memory technologies complementing each other or working side-by-side. A central question in these scenarios is how to place application data structures on the available memories to achieve optimal performance. Manual allocation is prohibitive for applications with more than an handful of memory objects. In this work, we present an algorithm for data placement on hybrid-memory system based on a set of memory allocation rules and global data placement decision algorithm. We propose a tool OracleHMS that implements our algorithm and provides recommendations on code change and memory setup to programmers. Our experiments on real hybrid-memory system show that our tool recommendations match or outperform manual optimization in a variety of scientific and data analytics applications.

Sun 18 Jun
Times are displayed in time zone: (GMT+02:00) Amsterdam, Berlin, Bern, Rome, Stockholm, Vienna change

14:00 - 15:30: ISMM 2017 - Session 3: Hybrid Memory Systems at Aula Master
Chair(s): Ben L. TitzerGoogle
ismm-2017-papers14:00 - 14:30
Mohammad DashtiUniversity of British Columbia, Alexandra (Sasha) FedorovaSimon Fraser University
ismm-2017-papers14:30 - 15:00
Ellis GilesRice University, Kshitij DoshiIntel Corporation, Peter VarmanRice University
ismm-2017-papers15:00 - 15:30
Ivy Bo PengKTH Royal Institute of Technology, Roberto GioiosaPacific Northwest National Laboratory, Gokcen KestorPacific Northwest National Laboratory, Stefano MarkidisKTH Royal Institute of Technology, Pietro CicottiSan Diego Supercomputer Center, Erwin LaureKTH Royal Institute of Technology