-
30. Re: JBoss EL performance vs Sun EL
matt.drees Sep 23, 2007 6:03 PM (in response to mgrouch)"fhh" wrote:
Well, if you don't have a profiler you should at least test EL performance only.
I agree."fhh" wrote:
Instead of building a tree of 1.000.000 components
Just wanted to point out that ui:repeat doesn't create a component per row, like c:forEach does. So his test probably only has a dozen or so components.
But I agree with your conclusion, nonetheless. -
31. Re: JBoss EL performance vs Sun EL
mgrouch Sep 23, 2007 9:49 PM (in response to mgrouch)I've opened JIRA to integrate performance test suite into EL build.
http://jira.jboss.org/jira/browse/JBSEAM-1925
"fhh" wrote:
Well, if you don't have a profiler you should at least test EL performance only.
My concern is overall response time. This is what I'm measuring.
I've done small changes to use cglib and I see that it got shorter. -
32. Re: JBoss EL performance vs Sun EL
christian.bauer Sep 23, 2007 11:58 PM (in response to mgrouch)My concern is overall response time. This is what I'm measuring.
How do you measure it? With a stopwatch? Seriously, it would take you 15 minutes to install a profiler and run whatever testcase you have. Maybe 30 minutes if you have to read the documentation before. -
33. Re: JBoss EL performance vs Sun EL
mgrouch Sep 24, 2007 12:21 AM (in response to mgrouch)Christian,
Please, take a look at previous messages in this thread.
It has all the information on how I measured it.
Additionally I did profiling and it displayed that the problem was in reflection.
You can reproduce it on you side. The current cost of calling getter via EL far
exceeds one via compiled code and I hope JBoss could address it.
You will probably need many profiler's snapshots while you try
to find best way to code around it, more than I'm able to provide you in this thread.
So I suggest you to do profiling on your side.
Thanks -
34. Re: JBoss EL performance vs Sun EL
amitev Sep 24, 2007 2:38 AM (in response to mgrouch)OFFTOPIC - could someone suggest me a profiler for eclipse?
-
35. Re: JBoss EL performance vs Sun EL
mustaghattack Sep 24, 2007 11:38 AM (in response to mgrouch)I'm also interested in profiling JBoss with eclipse.
- Tried the TPTP but I can't get it working (looks buggy to me)
- Tried the jboss-profiler (not integrated with eclipse, its a web app which uses logs), but I didn't find it that good (I can't see EL method call on the method call page, even when I specify include=* in the JVM arg).
So far I'm still looking for something ... -
36. Re: JBoss EL performance vs Sun EL
mgrouch Sep 24, 2007 12:20 PM (in response to mgrouch)I used JProbe with no problems
-
37. Re: JBoss EL performance vs Sun EL
mgrouch Sep 24, 2007 12:28 PM (in response to mgrouch)Here is a link with some info about fast reflection using cglib
www.sixlegs.com/blog/java/cglib-fastclass.html
FYI -
38. Re: JBoss EL performance vs Sun EL
mustaghattack Sep 24, 2007 12:32 PM (in response to mgrouch)About performance I'm wondering if JSF (+ Seam) is viable for a heavy loaded application (extranet).
I've downloaded Nuxeo which is the only open source "real" application I know that uses Seam. Ok it's designed for intranet usage but It's amazing how slow it is ...
I have a small server, for testing stuff. Trac (python app) is running very well on it.
I deployed a simple Seam app on that machine but the perf are not there : not even "usable" for a single user.
Even if I haven't done much optimization yet (at least the state saving is on the server side) I'm wondering how far I can improve this ...
Does anyone has done stress testing ? With which tools ? -
39. Re: JBoss EL performance vs Sun EL
mustaghattack Sep 24, 2007 12:48 PM (in response to mgrouch)Thanks, mgrouch.
I was looking for some open source software but I'll give a try to JProbe.
BTW thanks for all your performance investigation (++ for the performance tips thread). -
40. Re: JBoss EL performance vs Sun EL
mgrouch Sep 24, 2007 1:59 PM (in response to mgrouch)BTW
egrep "total|phase" server.log
will give you more insight on where (which JSF phase) the most time is
wasted
(server.log is JBoss server log) -
41. Re: JBoss EL performance vs Sun EL
vladimir.kovalyuk Sep 24, 2007 2:48 PM (in response to mgrouch)A hundred of interceptors are involved into the EL expression evaluation for each "enterprise_bean.property" construction.
I believe you need to "outject" your data into "request scope" before profiling in order not to mix EL and interceptors contribution into the result. -
42. Re: JBoss EL performance vs Sun EL
mgrouch Sep 24, 2007 2:57 PM (in response to mgrouch)I've tried outjection too already. Not much difference.
-
43. Re: JBoss EL performance vs Sun EL
mgrouch Sep 24, 2007 3:10 PM (in response to mgrouch)BTW:
Is there a rule of thumb when one should use
@BypassInterceptors
(or @Intercept(NEVER))
for best performance, without breaking the application?
Thanks -
44. Re: JBoss EL performance vs Sun EL
mgrouch Sep 25, 2007 6:28 PM (in response to mgrouch)One more offending function:
too many calls to the org.jboss.seam.Component.hasAnnotation method
also significantly degrade performance per our profiler snapshot
(although not for the code I gave in this thread).
It was also observed in this thread
http://www.jboss.com/index.html?module=bb&op=viewtopic&t=119573