Coupling on-line and off-line profile information to improve program performance

Abstract

In this paper we describe a novel execution environment for Java programs that substantially improves execution performance by incorporating both on-line and off-line profile information to guide dynamic optimization. By using both types of profile collection techniques, we are able to exploit the strengths of each constituent approach: profile accuracy and low overhead. Such coupling also reduces the negative impact of these approaches when each is used in isolation. On-line profiling introduces overhead for dynamic instrumentation, measurement, and decision making. Off-line profile information can be inaccurate when program inputs for execution and optimization differ from those used for profiling. To combat these drawbacks and to achieve the benefits from both online and off-line profiling, we developed a dynamic compilation system (based on JikesRVM) that makes use of both. As a result, we are able improve Java program performance by 9% on average, for the programs studied.