Date: Wed, 26 Feb 2003 09:46:36 +0100 From: Marc van Kempen <marc@bowtie.nl> To: java@freebsd.org Subject: performance results of 1.3.1 hotspot Message-ID: <3E5C7EEC.5000302@bowtie.nl>
next in thread | raw e-mail | index | archive | help
Hi, I just ran a performance test on the jdk 1.3.1 hotspot compiler with the latest patches released by Greg. We have developed a webbased content management system, running on tomcat 4.0.6. The CMS generates the webpages dynamically, as such it is pretty database intensive, however, the acl's (permissions) are cached and all database information is first loaded in java objects, then converted to xml with castor and then transformed to html using xslt. So we have a good mix of database operations and jvm operations. OS: FreeBSD 4.7-stable (27 Jan 2003) I ran apache benchmark (ab) on a webpage which is typical for the system. Document Length: 14604 bytes Concurrency Level: 4 and got the following times: - running jdk1.3.1 patchlevel 7 gives: Time taken for tests: 21.216 seconds Complete requests: 50 Failed requests: 0 Total transferred: 743450 bytes HTML transferred: 730200 bytes Requests per second: 2.36 Transfer rate: 35.04 kb/s received - running jdk1.3.1 patchlevel 8, with green threads (no hotspot): Time taken for tests: 20.311 seconds Complete requests: 50 Failed requests: 0 Total transferred: 743450 bytes HTML transferred: 730200 bytes Requests per second: 2.46 Transfer rate: 36.60 kb/s received - running jdk1.3.1 patchlevel 8, with hotspot enabled (I moved -server to the top in jvm.cfg). Time taken for tests: 62.054 seconds Complete requests: 50 Failed requests: 0 Total transferred: 772652 bytes HTML transferred: 758872 bytes Requests per second: 0.81 Transfer rate: 12.45 kb/s received (I ran this test several times, to make sure that the hotspot compiler got its chance) So the hotspot compiler jvm is approximately 3 times slower than the non-hotspot compiler with green threads. When I look at the profiling output in our logfile, I can see that after a while the cpu intensive parts gets faster (on average) than with the non-hotspot compiler, however the times vary much more. The database intensive part however is *much* slower, typically between 2000ms and 6000ms. For the non-hotspot compiler this is between 300ms and 1200ms. This is the puzzling part, the database performance should be about the same, and if anything, a little bit faster. The database we use is Solid 3.51, with a JDBC driver. The database runs on the same machine as the webserver. (How long does it take for hotspot to compile the neccesary parts of the code? Now I ran this test for about 10 minutes before doing the final test run) - running jdk1.3.1 patchlevel 8, with hotspot disabled, (I moved -classic to the top in jvm.cfg) just hangs. [no results] Regards, Marc. -- ---------------------------------------------------- Marc van Kempen tel. +31 40 2 64 98 60 BowTie Technology fax. +31 40 2 64 98 61 Raiffeisenstraat 7 mailto:marc@bowtie.nl 5611 CH Eindhoven http://www.bowtie.nl ---------------------------------------------------- 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?3E5C7EEC.5000302>