Identification and Quantification of Hotspots in Java Grande Programs

Abstract

In this paper we present a platform independent analysis of the dynamic profiles of the the Java Grande Forum benchmark suite. A useful platform independent metric for analysing eventual running time is to use a cost center design pattern to “charge” each bytecode to the appropriate method. This necessitates accessing an item in a hash table dictionary each time round the interpreter inner loop and incrementing a counter. The measurements from the experiment are significantly different for either static or dynamic method frequencies and show that for the Grande programs in the Java Grande Forum Benchmark suite 66% of the execution time, as measured by bytecode use, is spent in the top two methods.