karaf-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Christian Schneider <ch...@die-schneider.net>
Subject Re: Proposal - Lightweight standalone remote OSGi implementation for karaf-cellar
Date Tue, 16 Feb 2016 19:27:54 GMT
I did some tests with plain CXF some time ago. See
http://liquid-reality.de/display/liquid/2014/03/28/Revisiting+JMS+performance.+Improvements+in+CXF+3.0.0

So I would expect to get a peak performance of about 4000 calls per 
second or the double amount if one way is good enough.
For the JMS transport I have measured similar values.

I would expect CXF DOSGi to be about as fast a plain CXF as there is 
almost no overhead.

Probably you already know this but this is what I found about 
performance tests in java:
If you do performance tests then you have to make sure to warm up your 
system.
Actually the best way I found is to do a certain number of calls and 
then print the calls/s.
Do this in a loop and watch the numbers.

You will see that it starts quite slow. Then at some point it gets even 
slower and after that it is a lot faster.
This is when the hotspot compiler kicks in and optimizes the code. You 
have to make sure that you only
measure after that point.

Christian

On 16.02.2016 20:13, Johannes Utzig wrote:
> cschneider wrote
>> Just out of curiosity how many calls per second did you achieve using
>> RMI and the redhat impl?
>> I only tested CXF calls till now and wonder what difference the binary
>> format makes.
> The test was always done with 5 threads firing 1000 calls each, so 5000
> calls.
> The test setup was karaf 4.0.4 on 2 different desktop machines connected to
> the same switch.
>
> I don't have the exact numbers available right now, but our results for the
> different implementations where roughly:
>
> RMI - Baseline
> 1400ms
>
> Amdatu
> 8x slower
>
> ECF (generic)
> 10x slower
>
> CXF (SOAP transport)
> 2x slower
>
> Fabric d-OSGi
> same performance as RMI
>
> CXF showed very promising performance results, but seems to be a major
> hassle to deploy on JBoss.
> The redhat impl worked easily in both JBoss and Karaf and had the best
> performance results (5000 calls in less than 1.5 seconds). Please mind that
> the amount of parallel executions can have a big impact on the results
> depending on which remote osgi impl is used, but in this configuration the
> redhat impl produced twice the throughput of CXF.
>
> It is still a very rough implementation but if you are curious to try it
> yourself:
>
> https://github.com/seeburger-ag/fabric8/tree/resurrect/fabric/fabric-dosgi
>
> Best regards,
> Johannes
>
>
>
>
> --
> View this message in context: http://karaf.922171.n3.nabble.com/Proposal-Lightweight-standalone-remote-OSGi-implementation-for-karaf-cellar-tp4045343p4045400.html
> Sent from the Karaf - Dev mailing list archive at Nabble.com.


-- 
Christian Schneider
http://www.liquid-reality.de

Open Source Architect
http://www.talend.com


Mime
View raw message