SC²S Colloquium - July 13, 2012

From Sccswiki
Jump to navigation Jump to search
Date: June 14, 2012
Room: 02.06.020
Time: 3 pm, s.t.

Peter Strazdins: Performance Analysis of Large-scale Simulations on Supercomputers and Clouds

Simulations resulting from the high resolution of time-evolving physical systems require large-scale parallel processing for their timely execution. The effectiveness of this is dependent on good scalability of the resulting codes. Efficient and accurate performance analysis techniques are required to gain an clear understanding of scalability effects.

In this talk, I will describe some experiences in this area, with a global atmospheres simulation (Met Office Unified Model) and an electro-physical cardiac (Chaste) simulations, under highly memory-intensive benchmarks. This includes methodologies to efficiently analyze the performance characteristics of long-term simulations, and techniques and tools that can be used to gain insights into scalability and other performance issues.

First, I will describe the experimental results and accompanying analysis are on a supercomputer. Due to issues such a limited access to such a `premier facility', private and public clouds might be desirable for such applications. The remainder of the talk will describe an approach to easily transfer the applications together with their software stacks onto a cloud environment, and extend the performance analysis to these platforms.


Since 1990, Peter has been with the now Research School of Computer Science at Australian National University. He was closely involved with the ANU-Fujitsu CAP Parallel Computing Project over the years 1990 - 2002, and Since then, he has been involved in collaborative projects in the area of High Performance Computing with Sun Microsystems Laboratories, Intel Corporation and Platform Computing, and more recently Fujitsu Laboratories Europe.

His research interests include parallel numerical algorithms and applications, computer architecture and operating systems for high performance computers, middleware for cluster and cloud computers and computer simulation and performance analysis.