Mobicents JAIN SLEE 1.2.2 GA soak test

[Luis Barreiro]

Hi all,


I performed during the last few days a soak test on Mobicents Jain-SLEE 1.2.2.GA

It was supposed to handle 60 million calls at a call rate of 200 caps (It took more than 80 hours to complete), acting as UAS.

There were 83 calls that failed, all due to timeout, and 8 calls that stayed open at the end of the test (they were not finished by SLEE).
 
You can check the results for yourself in the attached report.pdf file.


Regards, Luis Barreiro.



The test was done under the following conditions:

Host: perf08.qa.atl.jboss.com

Processor: 2 x (Dual-Core AMD Opteron(tm) Processor 2220 SE, 2800.436 MHz, cache size: 1024 KB)

[lbarreir@perf08 ~]$ cat /proc/meminfo
MemTotal:      8234060 kB

[lbarreir@perf08 ~]$ uname -r
2.6.9-42.0.10.ELsmp

[lbarreir@perf08 ~]$ $JAVA_HOME/bin/java -version
java version "1.6.0_07"
Java(TM) SE Runtime Environment (build 1.6.0_07-b06)
Java HotSpot(TM) 64-Bit Server VM (build 10.0-b23, mixed mode)

JAVA_OPTS=" -Xms4000m -Xmx4000m -XX:+UseConcMarkSweepGC -XX:+CMSIncrementalMode -Dsun.rmi.dgc.client.gcInterval=3600000 -Dsun.rmi.dgc.server.gcInterval=3600000"

Mobicents Jain-SLEE 1.2.2.GA with Sip11 RA and Sip11-Loadtest-With-Dialogs example from SVN. All logs were disabled.

Load generator: SIPp 3.0, using the scenario file attached, started with the command:
./sipp 127.0.0.1:5060 -s yousendbye -sf uac.xml -i 127.0.0.1 -p 5050 -r 200 -l 100000 -m 100000000 -trace_stat -fd 1 -nd

[Eduardo Martins]

Are those tests made with the Event Router tunned regarding uncommitted AC attaches/detaches?

-- Eduardo

[Luis Barreiro]

There were no optimizations, besides the ones mentioned below (logs disabled, incremental GC, bigger heap, etc.).

AFAICT that change requires a new binary, compiled from source.

[Eduardo Martins]

Can you do it, since for a performance test we should setup the server to be as fast as possible? Not that I'm expecting a major increase, but for sure our customers may use specific configured builds...

-- Eduardo

[aayush]

Hi,

I am trying to replicate this soak test explained above.

After giving the options JAVA_OPTS=" -Xms4000m -Xmx4000m, and then
running JBOSS, the app server stops and gives an error like this:

Exception in thread "CompilerThread1" java.lang.OutOfMemoryError:
requested 952568 bytes for Chunk::new. Out of swap space?

Due to this, i am unable to replicate this soak test as per my
satisfaction and expectations.

I am doing this test on JBOSS 4.2.2 GA, SUN Solaris 10 OS, T 1000 (16
GB RAM). I am using a 64 bit server VM.

However, the JBOSS binary i am using is compiled on a 32 bit Linux OS.
Is this the reason, why i cant use a heap size in excess of 4 GB ?
(around 2^32)

Looking at the Mobicents soak test arguments given above -Xms4000m -
Xmx4000m, it seems that you are able to utilize a larger heap in
excess of 4 GB at runtime. Is the JBOSS being used in the QA labs
compiled on a 64 bit OS ?

Thanks in advance
Aayush

On Dec 10 2008, 9:30 pm, Luis Barreiro <lbbbarre...@gmail.com> wrote:
> There were no optimizations, besides the ones mentioned below (logs disabled, incremental GC, bigger heap, etc.).
> AFAICT that change requires a new binary, compiled from source.
> Eduardo Martins wrote:Are those tests made with the Event Router tunned regarding uncommitted AC attaches/detaches?

> -- EduardoOn Wed, Dec 10, 2008 at 4:27 AM, Luis Barreiro<lbbba...@gmail.com>wrote:Hi all,

> I performed during the last few days a soak test on Mobicents Jain-SLEE1.2.2.GA
> It was supposed to handle 60 million calls at a call rate of 200 caps (It took more than 80 hours to complete), acting as UAS.
> There were 83 calls that failed, all due to timeout, and 8 calls that stayed open at the end of the test (they were not finished by SLEE).
>
> You can check the results for yourself in the attached report.pdf file.
> Regards, Luis Barreiro.
> The test was done under the following conditions:Host:perf08.qa.atl.jboss.com
> Processor: 2 x (Dual-Core AMD Opteron(tm) Processor 2220 SE, 2800.436 MHz, cache size: 1024 KB)
> [lbarreir@perf08 ~]$ cat /proc/meminfo
> MemTotal: 8234060 kB
> [lbarreir@perf08 ~]$ uname -r
> 2.6.9-42.0.10.ELsmp
> [lbarreir@perf08 ~]$ $JAVA_HOME/bin/java -version
> java version "1.6.0_07"
> Java(TM) SE Runtime Environment (build 1.6.0_07-b06)
> Java HotSpot(TM) 64-Bit Server VM (build 10.0-b23, mixed mode)
> JAVA_OPTS=" -Xms4000m -Xmx4000m -XX:+UseConcMarkSweepGC -XX:+CMSIncrementalMode -Dsun.rmi.dgc.client.gcInterval=3600000 -Dsun.rmi.dgc.server.gcInterval=3600000"

> Mobicents Jain-SLEE1.2.2.GAwith Sip11 RA and Sip11-Loadtest-With-Dialogs example from SVN. All logs were disabled.

> Load generator: SIPp 3.0, using the scenario file attached, started with the command:

> ./sipp127.0.0.1:5060-s yousendbye -sf uac.xml -i127.0.0.1-p 5050 -r 200 -l 100000 -m 100000000 -trace_stat -fd 1 -nd

[Luis Barreiro]

Hello Aayush,

There is nothing wrong in using a 3 or 3.5 Gb heap.
In order to use heaps larger that, usually a 64bit JVM is enough.
The JBoss used was the usual 4.2.3 binary distribution.

Regarding the OutOfMemoryError, you can try to increase Permanent generation XX:MaxPermSize=512m,
although I'm not sure if it solves the problem.

I am also very curious about the performance of mobicents  on Solaris.
I may try it on our lab too, once the GA is out.

Regards Luis Barreiro.

[aayush]

Okay
Thanks Luis. I will try out different parameters and keep you posted.