Interactive Performance Analysis with Vampir

Date and Time: 
2013 Wednesday, April 3
CG1 Auditoriums
Andreas Knüpfer

Authors: Andreas Knüpfer, Thomas William

This presentation will introduce Vampir's interactive performance analysis for parallel applications based on event traces. It will provide an overview over the interactive visualization of event traces offered by Vampir including the support for the most important ways of parallelization in HPC as of today (MPI, OpenMP, CUDA, OpenCL, and more) as well as additional data sources like performance counters. Furthermore, it will show VAMPIR's scalability to tens and hundreds of thousands of processes or threads. Finally, typical performance weaknesses in HPC programs are discussed and how they manifest in the Vampir displays.

Speaker Description: 

Andreas Knüpfer has a MD in Maths and a Ph.D. in Computer Science. He currently is senior scientist at the Center for Information Services and High Performance Computing (ZIH) at TU Dresden (Germany). Research interests: parallel programming paradigms and parallel performance analysis

Thomas William received his Dipl.-Inf. degree in computer science from the TU-Dresden in 2007 for developing an I/O-tracing library for the D-Grid job-monitoring framework. He has over 7 years of experience in the benchmarking area and is especially interested in parallel I/O benchmarks. His actual research position at the Center for Information Services and High Performance Computing (ZIH) involves the integration of parallel analysis and visualization tools.

PDF icon vampir_e.pdf8.74 MB
Video recorded: 

Event Category: