Designing for Heterogeneity and Unreliability: Examples from Adaptive Mesh Refinement

Event Sponsor: 
Mathematics and Computing Science Seminar
Start Date: 
Jul 9 2015 - 10:30am
Building/Room: 
Building 240/Room 1406-1407
Location: 
Argonne National Laboratory
Speaker(s): 
Anshu Dubey
Speaker(s) Title: 
Computer Science Interviewee
Host: 
Marc Snir

High performance scientific software has many unique and challenging characteristics.  These codes typically consist of different stages of computation. The stages in turn use disparate algorithms and components that often put conflicting demands on the programming models and data structures. Serious performance and portability issues arise from heterogeneity of both algorithms and platforms. One can exploit abstractions to achieve portability, but for that the software must be designed to allow efficient interplay between various abstractions. Additionally, because of expected unreliability of future platforms, resiliency has become a critical element in software design. The structure of the application can be exploited to formulate a resiliency strategy that is adjustable depending on different computing environments. I have been recently working on the design of a software framework and a customizable resiliency strategy in the context of adaptive mesh refinement (AMR). A presentation of this work will form the first part of my talk. In the second part I will give a broad overview of my earlier work with software engineering of FLASH, a multi-component & multi-physics community code.

Miscellaneous Information: 

Bio:
Anshu Dubey is a member of the Applied Numerical Algorithms Group at the Computational Research Division of Lawrence Berkeley National Laboratory.  Before joining LBL she was the Associate Director of the Flash Center for Computational Science at the University of Chicago.  While at the Flash Center, she was the software architect for FLASH, a multi-physics multi-disciplinary scientific code with a wide user base. She received her PhD in Computer Science from Old Dominion University and B.Tech in Electrical Engineering from Indian Institute of Technology, New Delhi, India. Her research interests are in parallel computing, software architecture/engineering for high performance scientific computing, and AMR technologies.