The use of parallel/distributed programming increases as it enables high performance computing. There are many tools that help a user in the performance analysis of the application, and that allow to improve the application execution. As there is a high demand of computational power, new systems, such as large scale computer clusters, have become more common and accessible to everyone to solve complex problems. However, these systems generate a new set of problems related to the scalability of current analysis and tuning tools. Our automatic and dynamic tuning environment MATE does not scale well because it has a set of common bottlenecks in its architecture, and hence we have decided to improve the tool for providing dynamic tuning on large scale systems too. For this purpose, we are designing a new tool that introduces a tree-based overlay network infrastructure for scalable metrics collection, and to substitutes the current centralized performance analysis by a distributed one, in order to provide better scalability.
Weitere Kapitel dieses Buchs durch Wischen aufrufen
Bitte loggen Sie sich ein, um Zugang zu diesem Inhalt zu erhalten
Sie möchten Zugang zu diesem Inhalt erhalten? Dann informieren Sie sich jetzt über unsere Produkte:
- MATE: Toward Scalable Automated and Dynamic Performance Tuning Environment
- Springer Berlin Heidelberg
ec4u, Neuer Inhalt/© ITandMEDIA