A Connection Oriented Framework for Effectively Using Multicore in the Enterprise

Enterprise applications are built for client server architectures and rebuilt using service-oriented principles. When such web service based applications are hosted on multicore there are specific performance issues due to spatial and temporal characteristics of the multicore. Existing literature addresses the spatial case where inherent parallelism is exploited and the temporal case, where data is not available on main memory. We address a second temporal case, where the data is available in memory, but thread local data is of dissimilar sizes. We present a connection oriented framework and a testbed implementation of an enterprise application. We modeled this enterprise applications for both client-server as well as webservice configurations. We used Cache miss as a key metric for measuring spatial and temporal performance and then demonstrate how an "compute infrastructure awareness" using feedback can improve the performance on a hybrid multicore without modifying the application.