Hi Eric,
Yes I removed the registry lookup for the pre-mediate in synapse and now
the performance should be improved. Basically registry lookup is
affecting the performance badly.
Please have a look at the trunk.
Thanks,
Ruwan
Hubert, Eric wrote:
Hi all,
I have some updates. Some good and also some not that good messages. Please let me start with the positive one. This issue is also fixed on Windows. I really did a lot of testing and could not reproduce the problem. No matter how much I increased the load etc. So just looking at this, it is a great success!
While running my test cases I had the impression/feeling, that the cpu load on the esb machine has increased while the TPS has decreased. This was just a first impression, so I started some basic tests... Of course this is no reference environment and my aim was not to take absolut numbers. I just wanted to compare WSO2 ESB 1.6, which I had benchmarked a lot, with the snapshot version v5 + "the transport patch" (something near trunk I guess).
So I copied my synapse.xml to an old 1.6 version on the same machine and just changed the registry-config lines.
In my test I used the same helloWorld service I had sent to you together with soapUI.
I used a load test with strategy thread, 1 starting thread and 30 end threads for 120 seconds. I run the test once to warm up the environment. Then I re-run the same test and monitored the CPU load on the soapUI/client as well as o the ESB machine and the ESB log-file (all Windows in that case).
Here are my observations and the test results:
Between 1 and 5 threads the CPU load on the ESB was around 80 % or below, it then increased steadily. Having reached about 10 threads the CPU load was between 90 and 100% this didn't changed until the end of the test and 30 threads were reached.
The CPU load on the soapUI client (the JBoss was running on the same machine) started at around 40 % and around 85% at 10 threads. It then stayed between 90 and 100% until the end of the test.
No messages/exceptions on the ESB log.
soapUI results:
15210 requests processed, 140 tps, max response time 693 ms
I than stopped this instance and started the current version and repeated the same test (also including a warm-up):
During the warm-up I had no error on the ESB log. I then started the final
test. The CPU load on the ESB immediately started with more than 90% and ramped
up to 100% at around 4 Threads and stayed there at 100% consistently until the
end of the test. CPU load on the soapUI was rather low and peaked at about 50%.
One exception poped up on the console:
2008-05-07 22:46:51,629 [192.168.1.2-r2] [HttpServerWorker-18] ERROR ServerHandl
er Unexpected HTTP protocol error : Transfer-encoding header already present
org.apache.http.ProtocolException: Transfer-encoding header already present
at org.apache.http.protocol.ResponseContent.process(ResponseContent.java
:67)
at org.apache.http.protocol.BasicHttpProcessor.process(BasicHttpProcesso
r.java:304)
at org.apache.synapse.transport.nhttp.ServerHandler.commitResponse(Serve
rHandler.java:237)
at org.apache.synapse.transport.nhttp.ServerWorker.run(ServerWorker.java
:215)
at java.util.concurrent.ThreadPoolExecutor$Worker.runTask(Unknown Source
)
at java.util.concurrent.ThreadPoolExecutor$Worker.run(Unknown Source)
at java.lang.Thread.run(Unknown Source)
soapUI results:
7605 requests processed, 80 tps, max response time 1880 ms
So this doesn't look that good. Now this big performance degredation can have many sources. It doesn't have to be related to your last patch. It maybe also related to the pre-mediate registry lookup or other changes affecting performance. Unfortunately without your patch my testcase doesn't work at all, so I can't test this. But the pre-mediate issue has been fixed today or yesterday. I've seen some commit messages on the synapse dev list. So I will try to take another few minutes to build synapse/wso2 esb from source and test trunk.
Tomorrow I'm going to repeat the same test in our test lab with Linux machines. If I see the same problems, I will file a new issue regarding the performance degradation from 1.6 to 1.7.
I hope we can also resolve this problem. I will help as much as I can.
Regards,
Eric
________________________________
Von: [EMAIL PROTECTED] im Auftrag von Hubert, Eric
Gesendet: Mi 07.05.2008 18:39
An: [email protected]
Betreff: RE: AW: AW: AW: AW: AW: AW: [esb-java-dev] AsynchronousCloseException
Hi all,
we now can confirm that the problem is solved on Linux. Will do some
testing on Windows in a few hours.
Good work!
Regards,
Eric
_______________________________________________
Esb-java-dev mailing list
[email protected]
http://wso2.org/cgi-bin/mailman/listinfo/esb-java-dev
------------------------------------------------------------------------
_______________________________________________
Esb-java-dev mailing list
[email protected]
http://wso2.org/cgi-bin/mailman/listinfo/esb-java-dev
_______________________________________________
Esb-java-dev mailing list
[email protected]
http://wso2.org/cgi-bin/mailman/listinfo/esb-java-dev
