Grid Performance Prediction: Requirements, Framework, and Models

Grid performance prediction (GPP) is critical for grid middleware component services and grid (middleware) users to make optimized grid resource usage decisions to meet QoS requirements committed in SLAs. Other sophisticated concepts in grid like multi-criteria scheduling, grid capacity planning, and grid advance reservation are also dependant on GPP. GPP spreads over all levels of grid resources in multiple dimensions. We explore it at different coarse grain levels to discover the most wanted grid performance prediction needs of different human and service clients of grid and/or grid services, and present a comprehensive grid performance prediction framework to support at fine grain levels. We also provide a broad architecture of ALP (application level prediction) and NLP (network level prediction) for our grid middleware