An Autonomic Performance Environment for Exascale

摘要: Exascale systems will require new approaches to performance observation, analysis, and runtime decision-making optimize for efficiency. The standard "first-person" model, in which multiple operating system processes threads observe themselves record first-person profiles or traces offline is not adequate capture interactions at shared resources highly concurrent, dynamic systems. Further, it does support mechanisms adaptation. Our approach, called APEX Autonomic Performance Environment eXascale, provides sharing information among the layers of software stack, including hardware, systems, application code, both legacy. measurement components share across layers, merging data sets with collected by third-person tools observing hardware states node global-levels. Critically, a policy engine designed guide adaptation make algorithmic changes, re-allocate resources, change scheduling rules when appropriate conditions occur.