Virtual Institute — High Productivity Supercomputing


BMBF (2011 - 2014)

Performance dynamics of massively parallel codes (LMAC)

The time-dependent behavior of parallel simulation codes -- in particular when adaptive algorithms are employed -- is often irregular, making the understanding of performance dynamics an essential prerequisite for program optimization. While existing performance analysis tools typically provide detailed information along spatial dimensions like processes and nodes, performance dynamics has so far been a neglected aspect.

To support developers of parallel simulation codes with optimizing their codes, the LMAC project aims to extend the established performance analysis tools Vampir, Scalasca and Periscope with new functionality to automatically examine performance dynamics. In addition, the University of Oregon, associated partner, complements the project with corresponding extensions to the performance tool TAU. In a broader interpretation of "performance dynamics", the project also addresses performance differences across multiple program versions created in the optimization cycle. Overall, these measures lead to faster program optimization cycles with expected savings both in terms of program runtime and development effort. Rigorous quality assurance and extensive user documentation ensure that the proposed extensions to the tool environments can be deployed on the target systems within the project lifetime in production-level quality.

To address a broad user base both within the Gauss-Alliance and beyond, most of the performance tools are released free of charge to the community under an open-source license. Only Vampir, due to its complex user interface, is distributed commercially. The software products are accompanied by training- and support offerings through VI-HPS, and are expected to be maintained beyond the lifetime of the LMAC project itself.


Associated Partner


Promoted by Bundesministerium für Bildung und Forschung