LMAC - Performance Dynamics of Massively Parallel Codes

The time-dependent behaviour of parallel simulation codes is often irregular, especially when adaptive algorithms are employed. Therefore, understanding performance dynamics is an essential prerequisite for the optimization of such codes. To support developers in this task, the LMAC project (Leistungsdynamik massiv-paralleler Codes), funded under the 2nd BMBF call for "HPC software for scalable parallel computers", aims to extend the established performance analysis tools Vampir, Scalasca and Periscope with a new functionality allowing them to automatically examine performance dynamics. The project is coordinated by the German Research School for Simulation Sciences in Aachen and also involves the Jülich Supercomputing Centre, GNS mbH, RWTH Aachen, TU Dresden and TU Munich. In addition, the University of Oregon is an associated partner complementing the project with corresponding extensions to the performance tool TAU.

At the same time, the project will further develop and maintain the parallel instrumentation and measurement system Score-P, which forms the common base of all four tools mentioned above. Training and support for the software will be offered by VI-HPS, and will continue beyond the lifetime of the LMAC project itself. For more information, see the project website at http://www.vi-hps.org/projects/lmac.
(Contact: Christian Rössel, c.feld@fz-juelich.de)

from JSC News No. 202, 26 March 2012