In this paper we present our static analysis tool Iceberg that helps find potential performance bugs in concurrent Java programs by analyzing the code in critical sections. By look- ing at real world programs, we can start to understand what programmers are putting into critical sections and how that might affect performance. In this paper we look at a few spe- cific scenarios in which the code in a critical section might occasionally have high latency and slow down the overall execution of a program.

We look at 24 real world programs and present a num- ber of code metrics within critical sections that could indi- cate potential performance problems. These metrics include counts of variable-latency operations such as I/O and mem- ory allocation, as well as overall measurements of critical section size. Using our tool we are able to find critical sec- tions with unusual behavior compared to the other critical sections. We also present our future work on a dynamic analysis tool that takes the results of our static analysis tool and instruments the program to determine whether a true performance problem exists.