Performance comparison of Java (and .NET) runtimes

Mon Oct 18 17:52:00 GMT 2004

I have updated a web page showing results of performance measurement of
various Java runtimes.
 Performance Comparison of Java/.NET Runtimes (Oct 2004)
 http://www.shudo.net/jit/perf/
The benchmarks on the page are mainly compute intensive and not
server-side ones: SPEC JVM98, SciMark 2.0, Linpack benchmark and
Eratosthenes sieve. Such benchmark results seem not to be in much
demand these days but recently released JDK 5.0 pushed me into doing
performance measurement.
This time Opteron processor (with 64 bit and 32 bit binaries) was
tested in addition to Pentium 4. Another interesting point is .NET
Framework, which was tested with simple benchmarks ported to C#.
Possibly interesting points are:
 - C#/.NET performance
 - Startup process and/or JIT compilation may be heavier than
 Java runtimes including HotSpot Server VM.
 - Throughput of .NET Framework is pretty high.
 C#/.NET shown higher performance than HotSpot Server VM
 in Linpack benchmark with large data.
 - Difference between 64 bit (AMD64) code and 32 bit (x86) code on Opteron
 - 64 bit code is faster in some benchmarks, and in other
 benchmarks 32 bit code is faster.
 - 64 bit code should pay the penalty of larger pointer (reference)
 and benefit (a little ?) from faster 64 bit integer arithmetic.
Hope you enjoy,
 Kazuyuki Shudo	shudo@computer.org	http://www.shudo.net/