A methodology and a tool for parallel systems performance analysis

摘要

Describes a methodology and a tool for performance analysis of parallel systems and applications running on such systems. The premise is application-driven analysis of the hardware and software characteristics of parallel systems. The methodology and tool provide predictive capabilities either for the performance of a given application model problem or for the performance of the hardware/software system. The current approach is distinguished by two key components: scaling capabilities and hierarchical analysts capabilities.