[jboss-jira] [JBoss JIRA] Issue Comment Edited: (EJBTHREE-2042) Investigate performance issue related to use of @SecurityDomain on EJB

[Maros Bajtos (JIRA)]

    [ https://jira.jboss.org/browse/EJBTHREE-2042?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12567351#comment-12567351 ] 

Maros Bajtos edited comment on EJBTHREE-2042 at 12/3/10 3:32 AM:
-----------------------------------------------------------------

I've added blocking hotspots from my yesterday's profiling. Please see also my thread about terrible performance with concurrent ejb calls (which is more descriptive) - http://community.jboss.org/message/574096 . To sum it up, I've found out that if there is lot of (hundreds) concurrent ejb calls, they'll start to block each other on ClassLoader.loadClass method which is called from hotspots I've attached. ClassLoader.loadClass is synchronized method - so only one thread at given time may be executing it. Sample trace calling it:

org.jboss.ejb3.security.RoleBasedAuthorizationInterceptorv2.invoke
SecurityHelperFactory.getEJBAuthorizationHelper
SecurityActions.loadClass
ClassLoader.loadClass

The application I profiled is composed of lot of ejbs and most of the ejb calls are really fast - therefore with one request from web, there is maybe hundreds fast ejb calls - as soon as more users start to use application, these small concurrent ejb calls start to block each other because of synchronized method.

      was (Author: maros.bajtos):
    blocking hotspots
  
> Investigate performance issue related to use of @SecurityDomain on EJB
> ----------------------------------------------------------------------
>
>                 Key: EJBTHREE-2042
>                 URL: https://jira.jboss.org/browse/EJBTHREE-2042
>             Project: EJB 3.0
>          Issue Type: Task
>            Reporter: jaikiran pai
>            Assignee: jaikiran pai
>         Attachments: hotspots.tar.gz, ProfilingHotSpots.zip
>
>
> A user has reported that the use of @SecurityDomain shows a relatively poor performance (even when a @PermitAll) is applied to the bean. See the referenced forum thread for details. 
> Although, it can't be guaranteed that the performance of a bean method invocation with a @SecurityDomain will be same as that without a @SecurityDomain, the difference in the timings as reported in that thread looks a bit too high. This JIRA is to track to investigate the issue and see if there's any genuine performance bug.
>  

-- 
This message is automatically generated by JIRA.
-
For more information on JIRA, see: http://www.atlassian.com/software/jira