MuMMI integrates and extends PAPI, PowerPack, and Prophesy to facilitate the systematic measurement, modeling and prediction of performance and power in addition to allowing users to explore performance-power tradeoffs for parallel applications.
The major contributions of MuMMI include the following:
Extension of Prophesy's well-established performance modeling interface to encompass multicore and to incorporate power parameters and metrics into the performance models. A database component enables the modeling system to record and track relevant benchmark codes and results for characterizing multicore architectures.
Extension of PAPI's widely used hardware performance monitoring library to include the collection and interpretation of relevant data for various components of multicore systems, including cores, memory controllers, network fabrics, and thermal sensors. Building on top of PAPI's new multi-component architecture, MuMMI provides a user-configurable layer for defining derived higher level metrics relevant to the performance modeling task at hand and mapping these to available native events on a given platform.
Extension of an emerging power-performance measurement, profiling, analysis and optimization framework, called PowerPack, to multicore architectures. This work enables MuMMI to measure and predict power consumption at component (e.g. processor core) and function-level granularity in multicore architectures
Development of modeling and analysis techniques that can be used to explore the performance and power optimization space of multicore systems, especially targeting resource contention issues.
Development of a testbed and validation system for applying the measurement and modeling framework to a selected representative set of scientific applications.