Phase-based adaptive recompilation in a JVM

Abstract

Modern JIT compilers often employ multi-level recompilation strategies as a means of ensuring the most used code is also the most highly optimized, balancing optimization costs and expected future performance. Accurate selection of code to compile and level of optimization to apply is thus important to performance. In this paper we investigate the effect of an improved recompilation strategy for a Java virtual machine. Our design makes use of a lightweight, low-level profiling mechanism to detect high-level, variable length phases in program execution. Phases are then used to guide adaptive recompilation choices, improving performance. We develop both an offline implementation based on trace data and a self-contained online version. Our offline study shows an average speedup of 8.7% and up to 21%, and our online system achieves an average speedup of 4.4%, up to 18%. We subject our results to extensive analysis and show that our design achieves good overall performance with high consistency despite the existence of many complex and interacting factors in such an environment.

Efficient software-based online phase classification

“…In this paper we have focused on temporal phases [29,8], however, there are alternative definitions of phases such as code phases [15,12,18,13,32]. Instead of observing the execution behavior over time, the program binary is analyzed, where parts of program's control flow is grouped into phases.…”

“…This has been used to instrument code at the phase boundaries for various optimizations. Both software [12,13,32] and hardware [15,18] approaches have been suggested.…”

“…Gu and Verbrugge [13] used code phases to find the best optimization level for each function with respect to compilation/execution time for dynamic recompilation in Java virtual machines. Sondag and Rajan [32] instrumented phase boundaries for scheduling, and moved threads between heterogeneous processors when they changed phase.…”

“…A program phase is defined to be a period of execution with a stable behavior. There are several important optimizations, both offline [5,13,27,30,32] and online [8,15,17,24], which benefit from knowledge of program phases. For exam- ple, a CPU scheduler that knows the present phase of each runnable process can make better scheduling decisions and improve resource utilization [24,32].…”

“…For exam- ple, a CPU scheduler that knows the present phase of each runnable process can make better scheduling decisions and improve resource utilization [24,32]. Other example include dynamic recompilation where knowledge of program phases can help determining when it is worth recompiling [5,13,26], efficient profiling where profiling is only performed for a phase with yet unknown behavior [25], or collection of representative simulation points to speed up simulation [30]. The goal of this paper is to develop and explore online techniques for phase classification.…”