Date: Mon, 25 Sep 2000 22:26:41 +0900 From: SHUDO Kazuyuki <shudoh@muraoka.info.waseda.ac.jp> To: freebsd-java@freebsd.org Subject: Performance evaluation of JIT compilers Message-ID: <200009251326.WAA08721@cafe.muraoka.info.waseda.ac.jp>
next in thread | raw e-mail | index | archive | help
[I've posted this to Java-Linux ML, also. But many of BSD Java folks will be interested, so I'm writing.] Hi folks, I've measured performance of JIT compilers which work on Linux. If you're interested in, please see http://www.shudo.net/jit/perf/ Evaluated runtime systems are IBM JITC (in IBM JDK 1.3 and 1.1.8), HotSpot Client and Server VM in Sun JDK 1.3, JIT v3 in Kaffe 1.0.6, JBuilder Java 2 JIT by Inprise, OpenJIT 1.1.14, TYA 1.7v2, shuJIT 0.6.7 and an interpreter included in Blackdown JDK 1.2.2 FCS. I applied the following benchmark programs to above JITs and an interpreter: * SPEC JVM98 * SciMark 2.0 * Linpack benchmark * Eratosthenes Sieve References to these benchmarks and brief descriptions are available in the web page. Please keep in mind that these are only benchmarks and limited kinds of existing applications. A benchmark program can expose one surface of the runtime system. The SPEC JVM98 is a benchmark suite composed by real applications, so it can be a realistic good benchmark. But still the final composed score shown as `Geometric Mean' is affected very much by the composing method. The current composing method depends on the reference times (ex. 425 secs for javac). If the reference times are changed from the current values, the final scores also vary. Kazuyuki SHUDO Happy Hacking! Muraoka Lab., Grad. School of Sci. & Eng., Waseda Univ. To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-java" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200009251326.WAA08721>