19.num/JWD.stanley
.ls 2
.na
.LP
Performance Modeling and Prediction for Parallel Distributed Memory Machines

Ken Stanley 

(Professor J. W. Demmel)

(ARPA) DAAL03-91-C-0047

I am developing a methodology for producing fast scalable and portable
codes for distributed memory machines. To write such a
methodology, one must understand the performance of real applications
written in high-level languages and run on distributed memory parallel
machines, as well as everything that affects performance,
including algorithms, compilers, message passing systems, other
libraries, and hardware design of both the CPU and the interconnect
network. My methodology will help sophisticated users choose an
appropriate language, algorithm, compiler, message passing system and
architecture. In addition, my methodology will tell these users how
efficiently their algorithm is implemented, and whether their
implementation is performance-portable.