ode-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Mateusz Nowakowski (JIRA)" <j...@apache.org>
Subject [jira] Commented: (ODE-876) ODE 1.3.4 is 10 times slower than ODE 1.3.3
Date Mon, 22 Nov 2010 15:15:14 GMT

    [ https://issues.apache.org/jira/browse/ODE-876?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12934485#action_12934485
] 

Mateusz Nowakowski commented on ODE-876:
----------------------------------------

We've found what was the issue.

We did Java profiling and found that JaxpVariableResolver.getSimpleContent(...) run exponentialy
longer than the previous version.
See attached ODEProfiling.pdf

ODE-759 which was applied in 1.3.4 introduced  a change in org.apache.ode.bpel.elang.xpath20.runtime.JaxpVariableResolver
 that SaxonConfiguration is created each time JaxpVariableResolver.getSimpleContent(...) is
invoked, it was the cause of all evil.

We've applied simple patch (in ODEProfiling.pdf) and the results are even better than ODE
1.2.

Now I have two  requests:
- Include the fix and release 1.3.5.RC which is exactly 1.3.4 + only this fix.
Give us two weeks to bless 1.3.5.RC and release 1.3.5.
We've lost several months (and money) due to this issue and we can't wait another months to
some new release which can contains another bugs.

We would like to have it done ASAP.

- Introduce real word examples and real world performance tests.


> ODE 1.3.4 is 10 times slower than ODE 1.3.3
> -------------------------------------------
>
>                 Key: ODE-876
>                 URL: https://issues.apache.org/jira/browse/ODE-876
>             Project: ODE
>          Issue Type: Bug
>          Components: BPEL Runtime
>    Affects Versions: 1.3.3
>         Environment: ODE 1.3.4 JBI, Fuse 3.4.0.5, ODE database in memory
>            Reporter: Mateusz Nowakowski
>            Assignee: Rafal Rusin
>            Priority: Blocker
>             Fix For: 1.3.5
>
>         Attachments: ODE_performanceComparison.pdf, OdeProfiling.pdf, ping-pong-load-test-soapui-project.xml,
ping-pong-sa-1.0.0.zip
>
>
> We've recently migrated from ODE 1.2 JBI to ODE 1.3.4 JBI.
> We were unable to migrate earlier due to criticial bugs in JBI packages in 1.3.3 and
1.3.2..
> We ran the same performance test against new ODE 1.3.4 and previous ODE 1.2
> The same configuration, the same environment, the same test case, only different ODE
zip...
> For example:
> A service which has about 7 invokes bpel activities with ODE 1.3.4 spends on processing
5500 ms
> with ODE 1.2 600ms.. The timings spend on these 7 services doesn't change. JBI bus time
doesn;t change. Nothing has changed except ODE version.
> It happens the same for our all orchestrated services.
> The only thing which was changed is ODE version.
> I can't provide the test itself (prioprietary) but I can provide configuration.
> Our services are only in memory.
> ODE database is JBI external and it is Derby in memory. 
> However the results are shocking. We of course abandon ODE 1.3.4 ans stay with ODE 1.2,
but if it is not fixed, we may abandon ODE completely.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


Mime
View raw message