MSS soak test results

[Luis Barreiro]

Hi all,

Attached are the results for the sip-servlets soak test.

It ran on apache tomcat 6.0.14, JDK 1.5

Call rate was 100 calls/sec for 24h, for a total of 8.640.000 calls.

Regards, Luis Barreiro

[Luis Barreiro]

Hi all,

I repeated the MSS soak test, now using the incremental GC.

All other configurations remained the same (tomcat 6.0.14, jdk-1.5)
and the call rate was also the same (100 calls/sec for 24h, for a total of 8.640.000 call)

As expected, there are less pauses due to GC, which can be seen as less retransmissions and a smaller response time.

From both tests we can conclude that mss can hold this load for such a long period of time.
Both the number of failed calls and retransmissions are low.

Regards, Luis Barreiro

[Jean Deruelle]

Nice ! Thanks Luis,

I'm glad that we already got those figures on a to be released 0.4 version.
I will try to setup the Jboss performance test this week. Later on,
I'll try to setup also a Proxy test application.

Can it be setup to run periodically to ensure non regression ? If so
could you share the hudson link once this will be done so that I can
link to it from the product page on mobicents.org ?

By the way can you share the server hardware configuration and the jvm
options used for the test ?

Best regards
Jean

[Luis Barreiro]

Hi jean,

Yes, I can set up a hudson job for it, although the actual results won't be as good.
Still, we'll have a way to ensure that there are no regressions.

The test was done in the following conditions:

Hardware:
- Host: perf08.qa.atl.jboss.com
- Processor: 2x AMD 2220 SE
- RAM: 8GB

JVM:
- Version: Java HotSpot(TM) 64-Bit Server VM (build 1.5.0_12-b04)
- Options: -Xms4000m -Xmx4000m (-XX:+UseConcMarkSweepGC -XX:+CMSIncrementalMode )

[Jean Deruelle]

Great thanks

[Ivelin Ivanov]

This news made my day.
Would be great to setup a continues hudson run of the soak test and
include the proxy in the coverage.

[Vladimir Ralev]

Is this 100cps limit hardcoded in the test? May be it can do more?

[Luis Barreiro]

The test runs for the 24h with the same call rate, defined a the beginning.

Using throughput GC I got a maximum value of 120 call/sec. Above that some calls fail.
For a 24h test I go with a call rate lower than that.

CPU utilization was already at 80% / 90% so I guess you wont be able to go much further, at least for now.
I think we all agree that 100 calls/sec is good performance. Personally, it exceeded my expectations.

Regards, Luis Baarreiro.

[Jean Deruelle]

it was on UDP right ?
The goal would be to save a few days a month for performance and
improve it over every version.
But I think it's a good start too for a 0.3 version.