Publication Date

2009

Degree Type

Master's Project

Degree Name

Master of Science (MS)

Department

Computer Science

Abstract

The Java Virtual Machine (Standard Edition) normally interprets Java byte code but also compiles Java methods that are frequently interpreted and runs them natively. The purpose is to take advantage of native execution without having too much overhead for Just-In-Time compilation. A former SJSU thesis tried to enhance the standard policy by predicting frequently called methods ahead of their actual frequent interpretation. The project also tried to increase the compilation throughput by prioritizing the method compilations, if there is more than one hot method to compile at the same time. The paper claimed significant speedup. In this project, we tried to re-implement the previous work on a different platform to see if we get the same results. Our re-evaluation showed some speedup for the prediction approach but with some adjustments and only for server applications. It also showed some speedup for the prioritizing approach for all the benchmarks. We also designed two other approaches to enhance the original policy. We tried to reduce the start-up delay that is due to overhead of Just-In-Time compilation by postponing some of Just-In- Time compilation. We also tried to increase the accuracy of detecting frequently interpreted methods that contain nested loops. We could not gain any speedup for our former postponing approach but we could improve the performance using our latter measuring approach.

Download

Share

COinS