Efficient dynamic analysis of the synchronization performance of Java applications

Abstract

Concurrent programming has become a necessity in order to benefit from recent advances in processor design. However, implementing correct and scalable synchronization in concurrent code remains a challenge. Dynamic analysis of synchronization behavior is vital to determine where more sophisticated but error-prone synchronization pays off. We examine common approaches that developers use to identify and analyze concurrency bottlenecks in Java applications. We then describe key aspects of our ongoing research on a novel approach to Java synchronization analysis. Our approach provides developers with exhaustive information on the synchronization behavior of their application, but incurs such low overhead that it is feasible to use it for monitoring production systems. Unlike other methods, our approach can precisely show where optimizations have the largest impact.