MSS 0.6 performance tests

[Luis Barreiro]

Hi all,

I've been doing some performance tests with MSS 0.6.
The results are quite promising.

Attached are a report file generated from SIPp statistics.
It shows  MSS handling 1 million  calls  in one hour (280 cps)
I'll try to extend this test to 24h in a near future, maybe after next version comes out.

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=" -Xms6000m -Xmx6000m -XX:+UseConcMarkSweepGC -XX:+CMSIncrementalMode -Dsun.rmi.dgc.client.gcInterval=3600000 -Dsun.rmi.dgc.server.gcInterval=3600000"

MSS 0.6 with SimpleSipServlet example from SVN and all logs disabled.

Load generator: SIPp 3.0, using the scenario attached, started with the command:
./sipp 127.0.0.1:5080 -s yousendbye -sf uac.xml -i 127.0.0.1 -p 5050 -r 280 -l 100000 -m 1000000 -trace_stat -fd 1 -nd

[Vladimir Ralev]

Is there an option to record the CPU utilization during the test?

[Luis Barreiro]

Not directly from SIPp.

It is possible to get the CPU usage, along with other JVM parameters such as heap usage using JMX, but I didn't got into that.

Anyway, during the test I used the top command and saw  that even with higher call rates CPU is usually under 50%

[Luis Barreiro]

Attached is another SIPp report, done in the same conditions.

The test start with 1 call in the first second, then 2 on the second, 3 on the third ... and so on.
It ends after 100.000 calls, at a call rate just below 450 cps

MSS 0.6 handles all calls, without failures.
We can conclude that it can handle load peaks of more than 400 cps.

Regards, Luis Barreiro.

[Jean Deruelle]

Nice results :-)
It means we got close to a 3x boost perf improvement in 3 versions
(from 0.3 where we were handling 100 caps to 0.6 handling close to 300
caps) :-)

As for the tests against the next version, concurrency control will be
introduced in 0.7 and set to SipSession as a default.
To get max perf results you should revert to no concurrency control.

Thanks for running those tests Luis.

[Ivelin Ivanov]

Thanks, Luis.
Please link to this report from the Mobicents performance tuning page.
http://groups.google.com/group/mobicents-public/web/mobicents-performance-guide

[aayush bhatnagar]

Hello Luis, 

Are there any latest performance figures / test results available for SLEE just like for MSS ?

The only figures i have are those given in the FAQ section: http://www.jboss.org/files/jbosslabs/design/mobicents/html/faq.html

Any updates to these figures so far ? 

I was going through the internet for performance analysis of SLEE based platforms and applications. I found a white paper of SLEE performance figures but from another vendor : jNETx. 

Even though it is a SLEE implementation by jnetX and not Mobicents, but i am sure that the white paper will be an interesting read for everyone. Shows the potential of SLEE.

I am attaching it in this mail.

:) 

--
Aayush

[Luis Barreiro]

Hello Aayush,

Please don't be impatient, performance results for SLEE will come out
soon. I'm working on that.

Regarding the jNETx numbers, there is a fundamental difference with
ours:
In our case you have everything to replicate the test.

I won't make further comment about their tests.


Regards, Luis Barreiro.


On Oct 30, 5:31 am, "aayush bhatnagar" <abhatnagar192...@gmail.com>
wrote:

> Hello Luis,
> Are there any latest performance figures / test results available for SLEE
> just like for MSS ?
> The only figures i have are those given in the FAQ section:http://www.jboss.org/files/jbosslabs/design/mobicents/html/faq.html
>
> Any updates to these figures so far ?
>
> I was going through the internet for performance analysis of SLEE based
> platforms and applications. I found a white paper of SLEE performance
> figures but from another vendor : jNETx.
>
> Even though it is a SLEE implementation by jnetX and not Mobicents, but i am
> sure that the white paper will be an interesting read for everyone. Shows
> the potential of SLEE.
>
> I am attaching it in this mail.
>
> :)
>
> On Tue, Oct 28, 2008 at 3:41 AM, Ivelin Ivanov <
>
>
>
> ivelin.atanasoff.iva...@gmail.com> wrote:
>
> > Thanks, Luis.
> > Please link to this report from the Mobicents performance tuning page.
>

> >http://groups.google.com/group/mobicents-public/web/mobicents-perform...

>  jNetX_Performance.pdf
> 549KViewDownload

[Ivelin Ivanov]

Ditto!

Its hard to compare a coin that you hold in your hand to a coin that
you can only see on one side but not the other.

[aayush bhatnagar]

Thanks for the replies Luis, Ivelin

I was not comparing vendors here.

I found something interesting on JAIN SLEE, so shared it as light reading.
As was correctly said, its hard to comment on figures published by
others. Their target platform and deployment architecture is not
known, their SUT's application complexity is not known...and most
importantly, their benchmarking standards compliance is not known.

I am currently considering Tispan NGN as the basis for my tests.

Best Regards
Aayush.

--
Aayush