A Lightweight Approach of Automatic Resource Configuration in Distributed Computing

A key problem in executing performance critical applications on distributed computing environments (e.g. the Grid) is the selection of resources for execution. A lot of research related to "automatic resource selection" has been made to allocate the best-effort resources on behalf of users to optimize the execution performance. However most of current approaches are based on the static principle (i.e. resource selection is performed prior to execution) and need detailed application specific information. In the paper, we introduce a lightweight approach for automatic resource selection/configuration. This approach is based on a simple control theory: the application continuously reports performance values to the middleware Application Agent (AA), which relies on the reported values to decide how to dynamically reconfigure the execution environment during the execution to ensure userspsila performance requirements (e.g. execution deadline, running N iteration per second). We divide the research into two paradigms: neglecting network latency and considering network latency. For the first paradigm, we use a linear prediction with Kalman filter to find the expected RC to satisfy certain performance requirement. For the second, we let AA probe possible RCs and rollback the bad RCs, to look for a local optimized RC that can provide applications highest performance.