Adaptive optimization in the Jalapeno JVM

摘要: Future high-performance virtual machines will improve performance through sophisticated online feedback-directed optimizations. this paper presents the architecture of Jalapeno Adaptive Optimization System, a system to support leading-edge machine technology and enable ongoing research on We describe extensible architecture, based federation threads with asynchronous communication. present an implementation general that supports adaptive multi-level optimization purely statistical sampling. empirically demonstrate profiling technique has low overhead can startup steady-state performance, even without presence The also describes evaluates inlining edge is written completely in Java, applying described techniques not only application code standard libraries, but itself.