Kevin Huck

"Integrating Knowledge, Automation, and Persistence with PerfExplorer and PerfDMF"
University of Oregon


Abstract:

Parallel performance diagnosis and engineering in the future will require more sophisticated support for analysis automation, knowledge integration, intelligent problem discovery, and perscriptive feedback. How such capabilities are designed and implemented in a performance analysis framework includes consideration for programmability, extensibility, interoperability, and resuse. In this talk we look at this issues in the context of PerfDMF and PerfExplorer, and present recent applications. The PerfExplorer framework has evolved to include a scripting interface and inference engine to capture performance expert reasoning and build an analysis knowledge base for intelligent problem solving. We will discuss recent success analyzing code generated by the OpenUH compiler, with the eventual goal of providing feedback to the compiler to improve loop optimizations. We are also integrating with the CCA framework to provide runtime Computational Quality of Service (CQoS) functionality. We will discuss these integrations, the potential interfaces between the tools, and how PerfDMF data storage and PerfExplorer automation and inference can be exposed for more analysis applications.