SPEC JVM98 Benchmarks

 

Table: SPEC JVM98 benchmark codes used in the experiments.

Benchmark	Brief Description
compress	A high-performance application to compress/uncompress large files; based on the Lempel-Ziv method (LZW)
jess	A Java expert shell system based on NASA's CLIPS expert shell system
db	A small database management program that performs several database functions on a memory-resident database
javac	JDK 1.0.2 Java compiler
mpeg	MPEG-3 audio file compression application
mtrt	Dual-threaded ray tracer; the only multi-threaded application in the suite
jack	A Java parser generator with lexical analyzers; an early version of what is now called JavaCC

 

We use all seven applications from the SPEC JVM98 benchmark suite [27] for our experiments that are briefly described in Figure . The benchmark programs can be run using three different problem sizes, which are named as s100, s10 and s1. While the problem sizes are larger for the s100 dataset size, the sizes do not scale as designated by the labels 100, 10 and 1. In interest of simulation time, especially, the interpreted mode in s100 mode, we use the smaller problem size (s1). We believe that the s1 data set is representative of shorter running applications or applets. While some of the results from the s1 observations would be applicable to the larger data sets, it must be noted that the impact of garbage collection and dynamic compilation impacts tend to change for the larger data sets. Thus, we provide s10 and s100 energy breakdowns in Figure  to emphasize this difference.