Back to full Calendar

AMCS Seminar: Performance Modeling as the Key...

Start Date: December 03, 2011
End Date: December 03, 2011

Performance Modeling as the Key to Extreme Scale Computing

By

Dr. Bill Gropp (University of Illinois)

Parallel computing is primarily about achieving greater performance than is possible without using parallelism.  Especially for the high-end, where systems cost tens to hundreds of millions of dollars, making the best use of these valuable and scarce systems is important. Yet few applications really understand how well they are performing with respect to the achievable performance on the system.  The Blue Waters system, currently being installed at the University of Illinois, will offer sustained performance in excess of 1 PetaFLOPS for many applications. However, achieving this level of performance requires careful attention to many details, as this system has many features that must be used to get the best performance.  To address this problem, the Blue Waters project is exploring the use of performance models that provide enough information to guide the development and tuning of applications, ranging from improving the performance of small loops to identifying the need for new algorithms.  Using Blue Waters as an example of an extreme scale system, this talk will describe some of the challenges faced by applications at this scale, the role that performance modeling can play in preparing applications for extreme scale, and some ways in which performance modeling has guided performance enhancements for those applications.

Biography: William D. Gropp is Paul and Cynthia Saylor Professor of Computer Science at the University of Illinois Urbana-Champaign.  He is also the Deputy Director of Illinois' Research Institute for Advanced Computing Applications and Technologies. Gropp helped to create the Message Passing Interface, also known as MPI, and the Portable, Extensible Toolkit for Scientific Computation, also known as PETSc. Gropp was awarded the Sidney Fernbach Award in 2008 "for outstanding contributions to the development of domain decomposition algorithms, scalable tools for the parallel numerical solution of PDEs, and the dominant HPC communications interface." In 2009, Gropp received an R&D 100 Award for PETSc. In February 2010, he was elected to the National Academy of Engineering "for contributions to numerical software in the area of linear algebra and high-performance parallel and distributed computation." In March 2010, he was honored with the IEEE TCSC Medal for Excellence in Scalable Computing.

Date & Time:  Saturday, December 3rd, 2011 from 01:45 pm to 02:45 pm

 

Location:        Lecture Hall II # 2325,  building 9

 

Refreshments: Available in 2325 @ 01:30 pm

For more information:

Contact:
Email:
Phone:
Cell:
Website: