BLP: Applying ILP Techniques to Bytecode Execution

Abstract

The popularity of Java has resulted in a flurry of engineering and research activity to improve performance of Java Virtual Machine (JVM) implementations. This paper introduces the concept of bytecode-level parallelism (BLP)--data- and control- independent bytecodes that can be executed concurrently--as a vehicle for achieving substantial performance improvements in implementations of JVMs, and describes a JVM architecture--JVM-BLP--that uses threads to exploit BLP. Measurements for several large Java programs show levels of BLP can be as high as 14.564 independent instructions, with an average of 6.768.