In a benchmarking of our server software I found about a 34% speed improvement (if my math is correct) on using JDK 1.6.0_12 over JDK 1.5.0_17.
Filed under: Performance
February 12, 2009 • 6:19 pm
Ben J. Christensen
Software Development and Other Random Stuff
Twitter Updates
- Having been to Karachi twice and worked daily with Pakistanis for 7+ years, I agree => “Good Things About Pakistan” http://t.co/IUfjJ2n8 1 day ago
- "We don’t lack for potential solutions … so much as we do the willingness to intelligently sacrifice in the short term" http://t.co/fOMCwn0s 1 day ago
- Regarding solitude … “[many believe] that creativity and productivity comes from a gregarious place. This is nonsense.” http://t.co/NrIVtAYe 4 days ago
- "achievement is often the result … of doing something increasingly difficult until you get the result you seek" http://t.co/tgPvuVo2 4 days ago
- Tell Congress: Don’t censor the web! http://t.co/55zuPJ4f 2 weeks ago
Theme: Grid Focus by Derek Punsalan.
Can you provide the environment you tested in? What tests were run? Sample size, etc?
Machine:
Amazon EC2 Large Instance 7.5 GB of memory, 4 EC2 Compute Units (2 virtual cores with 2 EC2 Compute Units each), 850 GB of instance storage, 64-bit platform
http://aws.amazon.com/ec2/#instance
[root@domU-12-31-39-02-D9-01 mnt]# uname -a
Linux domU-12-31-39-02-D9-01 2.6.21.7-2.fc8xen #1 SMP Fri Feb 15 12:34:28 EST 2008 x86_64 x86_64 x86_64 GNU/Linux
[root@domU-12-31-39-02-D9-01 mnt]# cat /proc/cpuinfo
model name : Dual-Core AMD Opteron(tm) Processor 2218 HE
cpu MHz : 2599.998
cache size : 1024 KB
As for the tests, they are a proprietary application that I can’t provide, but it is a java client application running as a standalone java process connecting via RMI to a standalone java process acting as the server.
JVM arguments for the server are:
$JAVA_HOME/bin/java -server -Xms1g -Xmx6g -XX:MaxPermSize=128m -XX:+UseConcMarkSweepGC -XX:+CMSPermGenSweepingEnabled -XX:+CMSClassUnloadingEnabled -verbose:gc -XX:+PrintGCDetails -XX:+PrintGCTimeStamps -Xloggc:../logs/gc.log -Dsun.rmi.dgc.client.gcInterval=3600000 -Dsun.rmi.dgc.server.gcInterval=3600000 -Dcom.sun.management.jmxremote -cp $CLASSPATH APPLICATION_CLASS_HERE >> ../logs/console.log 2>&1 &
The client JVM args are:
$JAVA_HOME/bin/java -Xmx128m -cp $CLASSPATH APPLICATION_CLASS_HERE
Each request/response is ranging in size from 5kb to 50kb transported over RMI.
Each execution of the test executed 528 samples (the number is derived from a standard dataset used for the tests, hence the reason it’s not exactly 500 or something similar).
This is by no means a fully scientific test, but it is “real-world” for the application I was testing.
Ben
Great Ben for posting…Its very useful for myself…Thanks for posting…
nice…
Hey Nice comparison man!! But any idea why this subtle difference in JDK 1.6 in terms of performance?Is this comparison done for same code compiled with JDK 1.5 and JDK 1.6? If its same code then its really something interesting
It was the same code and compiled just once on JDK 5, not separately for each.