hStreams: a way forward to Exascale

Event Sponsor: 
Argonne Leadership Computing Facility Seminar
Start Date: 
Jun 1 2016 - 9:30am
Building/Room: 
Building 240/Room 1406-1407
Location: 
Argonne National Laboratory
Speaker(s): 
Chris Newburn
Speaker(s) Title: 
Intel
Host: 
Hal Finkel

Application developers would love to enjoy the benefits of heterogeneous computing elements and new memory kinds, but tend not to want to become experts in technologies and APIs in order to benefit from them.  hStreams is a library-based runtime that offers a higher-level abstraction to enable ease of use, but also full control over performance where desired.  This talk offers a vision for solving the problems of tomorrow, grounded in the hStreams architecture.  hStreams offers a uniform tasking interface for concurrency within and across different kinds of computing elements with a composable, dynamic scheduling framework.  It's relevant for scheduling across sub-clusters of nodes, coprocessors, GPUs, FPGAs and other such elements.  It also provides a declarative interface for new memory kinds and other properties.  This talk provides an overview of these technologies, as a starter for engagement in looking at ANL/ALCF's needs for such features.

Bio:

Chris (CJ) Newburn serves as an HPC architect, focused on Intel's Xeon Phi product family.  He has contributed to a combination of hardware and software technologies over the last twenty years.  He has a passion for making the richness of heterogeneous platforms easier to use.  He has over 80 patents.  He wrote a binary-optimizing, multi-grained parallelizing compiler as part of his Ph.D. at Carnegie Mellon University.  Before grad school, in the 80s, he did stints in a couple of start-ups, working on a voice recognizer and a VLIW mini-super computer.  He's delighted to have worked on volume products that his Mom uses.