Title:      ScalaExtrap: Trace-Based Communication Extrapolation for SPMD Programs

Speaker:     Frank Mueller, Professor, North Carolina State University

Time:        Dec. 13, 14:30-16:00

Venue:       FIT 1-315, Tsinghua University




Performance modelling for scientific applications is an important task

for assessing potential performance, bottlenecks assist in

hardware procurement for HPC systems.  Recent progress on

communication tracing provides novel opportunities for communication



We contribute a fundamentally novel approach. By extrapolating

communication traces ution time of a parallel application, a

set of traces for smaller number of nodes provides the base to

synthetically generate a trace for larger number of nodes.  The

extrapolated larger trace can subsequently be replayed to empirically

detect scalability problems or analyzed to detect communication

patterns or determine constraints on scalability, which is without




Frank Mueller (mueller@cs.ncsu.edu) is a Professor in Computer Science

a member of multiple research centers at North Carolina State

University. Previously, he held positions at Lawrence Livermore

National Laboratory Humboldt University Berlin, Germany. He

received his Ph.D. from Florida State University in 1994.  He has

published papers in the areas of parallel distributed systems,

embedded real-time systems compilers.  He is a member of ACM

SIGPLAN, ACM SIGBED a senior member of the ACM IEEE Computer

Societies.  He is a recipient of an NSF Career Award, an IBM Faculty

Award, a Google Research Award a Fellowship from the Humboldt