Online Feedback-Directed Optimizations for Parallel Java Code

Multi-core processors are nowadays standard in servers, desktop computers, and mobile devices. To make applications exploit the new hardware resources that are provided by the large number of cores, programmers must write parallel code that runs concurrently on the available cores. State-of-the-art parallel programming tech-niques require the programmer to define concurrent code regions and the conditions (concurrency condition) under which concurrent code regions are executed in paral-lel. If the concurrency conditions are set unlucky, the performance of parallel code can drop way below its potential. Moreover, it is even possible that parallel code performs slower than a corresponding sequential version, if the overhead that is as-sociated with every parallel execution exceeds the performance gain of a parallel execution. This thesis examines an automated, online, and feedback-directed approach to determine the concurrency conditions under which concurrent code regions are ef-fectively executed in parallel. The approach is automated, since the runtime syste