Authors: Markus Geimer
This talk will provide an overview of the open-source performance analysis toolset Scalasca. Scalasca features a highly scalable method to identify potential performance bottlenecks in parallel applications based on MPI and/or OpenMP, in particular those concerning communication and synchronization. We will introduce Scalasca's analysis methodology based on an automatic analysis of detailed event traces, demonstrate its scalability to a million concurrent threads of execution, and outline ongoing research. Furthermore, the CUBE analysis report explorer will be introduced, which is used for examining call-path profiles produced by Scalasca as well as the community instrumentation and measurement infrastructure Score-P covered by a separate talk.
After earning his Ph.D. in Computer Science from the University of Koblenz-Landau (Germany) in 2005, Markus Geimer joined the Jülich Supercomputing Centre as a research scientist beginning of 2006. Since then he is working on the Scalasca performance analysis toolset as the lead developer of its parallel trace analysis component. Furthermore, he is also heavily involved in training activities related to Scalasca as well as Score-P, and has published many refereed articles in journals and conference or workshop proceedings.
Attachment | Size |
---|---|
Geimer_Scalasca.pdf | 2.19 MB |