A VM Infrastructure for Understanding the Hardware Performance of Java Applications

Modern Java programs such as middleware and application servers include many complex software components. Improving the performance of these Java applications requires a better understanding of the interactions between the application, virtual machine, operating system, and architecture. Hardware performance monitors, which are available on most modern processors, provide facilities to obtain detailed performance measurements of long-running applications in real time. In this paper we describe the design and implementation of extensions to a virtual machine to access hardware performance monitors. The goal of the extended virtual machine is to enable a low overhead mechanism to collect threadspecifc, fine-grained temporal hardware performance information on a multiprocessor. We extend Jikes RVM, an open source research virtual machine for executing Java programs, to Periodically collect performance monitor information. We correlate the performance data to each of the application and VM threads and demonstrate the utility of the VM extensions by measuring the memory performance of SPECjbb2000 on a Power4 multiprocessor. We measure the load latency of each thread, separating the application threads from the VM threads, report the processor on which they execute, and show how the load latency varies over time for one application thread. We also discuss future extensions
to this infrastructure.