Today, Java is regularly used to implement large multi-threaded server-class ap- plications, whose scalability could be hampered by the execution of critical sections. Profiling is needed to identify critical sections that are problematic. However, profiling such applications is challenging, due to their long running times and the range of possible runtime conditions. We propose Free Lunch, a new profiler designed to identify locks and critical sections that hamper scalability. Free Lunch is designed around a new metric, critical section pressure, and can be used in-vivo, while the application is run by end-users. Using Free Lunch, we have identified a contention phase in the distributed Cassandra NoSQL database and in several applications from the DaCapo benchmark suite. On the latter, we were able to improve the performance of the Xalan benchmark by 15%. In an evaluation on over thirty applications, we found that the overhead of Free Lunch is never greater than 6%.
