axis-java-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Glen Daniels <g...@thoughtcraft.com>
Subject Re: AXIS-1423 : Peter's test case for performance
Date Tue, 20 Jul 2004 19:06:24 GMT
Hi Peter:

Hm - did you get a look at what the majority of the objects were?

--Glen

----- Original Message ----- 
From: "Peter Molettiere" <pietro@axonstudios.net>
To: <axis-dev@ws.apache.org>
Cc: "Tom Jordahl" <tomj@macromedia.com>; <dims@apache.org>; "Glen Daniels"
<gdaniels@sonicsoftware.com>
Sent: Tuesday, July 20, 2004 2:44 PM
Subject: Re: AXIS-1423 : Peter's test case for performance


>
> Well, interesting results from running the test case overnight. Heap
> size very slowly increased to over 200M after running from last night
> at about 6:00 to this morning at about 10:30.
>
> --Peter
>
> On Jul 19, 2004, at 6:19 PM, Peter Molettiere wrote:
>
> >
> > Ok, I'll add comments there. Just to be sure that the leak is gone,
> > I'm leaving the test case run overnight tonight, to see how it looks
> > in the morning.
> >
> > Also, it's a weird thing -- running my test case, I see it run for 15
> > or 30 minutes, then get OOM errors on the client, even being run with
> > -Xmx1024M.
> >
> > --Peter
> >
> > On Jul 19, 2004, at 5:44 PM, Davanum Srinivas wrote:
> >
> >> Peter,
> >>
> >> Thanks. FYI, please add your comments about the large mem needed for
> >> deser to the existing bug
> >> (http://nagoya.apache.org/jira/browse/AXIS-1391). I think that bug is
> >> the closest to your description.
> >>
> >> thanks,
> >> dims
> >>
> >> On Mon, 19 Jul 2004 17:32:38 -0700, Peter Molettiere
> >> <pietro@axonstudios.net> wrote:
> >>>
> >>> OK, I've spent most of the afternoon looking at the latest axis in
> >>> JProfiler, and it looks a lot better than it did before. Running the
> >>> test case I posted uses about 150M of heap on the server, mainly to
> >>> deal with serialization I think. Axis could definitely use some
> >>> improvement in the numbers of transient objects it creates during
> >>> this
> >>> process, but I'm guessing that's a fairly significant architectural
> >>> change.
> >>>
> >>> Anyway, back to my test results. Tomcat 5 seems to use about 10 to 20
> >>> fewer megs of RAM, although we run into other issues with axis on
> >>> Tomcat 5 with our app, so we didn't spend too much time looking
> >>> there.
> >>> Tomcat 4.1.30 uses about 12M after startup, and axis will burst up to
> >>> about 150M during serialization. Stopping the tester client, and
> >>> running gc (several times) on the server seems to release all the
> >>> memory it used.
> >>>
> >>> I think people using axis to send large messages are still going to
> >>> see
> >>> Out Of Memory Errors on a regular basis, though, due to the amount of
> >>> memory used during the serialization process. I'd even consider
> >>> filing
> >>> that as a bug -- although I'm not sure how that would be received.
> >>> Maybe the (in progress) streaming support will help that.
> >>>
> >>> --Peter
> >>>
> >>>
> >>>
> >>> On Jul 19, 2004, at 11:28 AM, Peter Molettiere wrote:
> >>>
> >>>>
> >>>> Cool... I'll check this version.
> >>>>
> >>>> --Peter
> >>>>
> >>>> On Jul 19, 2004, at 9:58 AM, Davanum Srinivas wrote:
> >>>>
> >>>>> ok. checked a version with commons' BOAS.
> >>>>>
> >>>>> -- dims
> >>>>>
> >>>>> On Mon, 19 Jul 2004 12:14:09 -0400, Davanum Srinivas
> >>>>> <davanum@gmail.com> wrote:
> >>>>>> Note that i may incorporate Nishant's suggestion of commons'
 BOAS
> >>>>>> and
> >>>>>> ask you to test that as well :)
> >>>>>>
> >>>>>> -- dims
> >>>>>>
> >>>>>> On Mon, 19 Jul 2004 08:58:36 -0700, Peter Molettiere
> >>>>>>
> >>>>>>
> >>>>>> <pietro@axonstudios.net> wrote:
> >>>>>>>
> >>>>>>> I'll run them this afternoon. Thanks for all your effort!
I'll
> >>>>>>> let
> >>>>>>> you
> >>>>>>> know how it goes once I'm at the machine I have my profiler
set
> >>>>>>> up
> >>>>>>> on.
> >>>>>>>
> >>>>>>> --Peter
> >>>>>>>
> >>>>>>>
> >>>>>>>
> >>>>>>> On Jul 17, 2004, at 12:56 PM, Davanum Srinivas wrote:
> >>>>>>>
> >>>>>>>> Peter, Nishant, Glen,
> >>>>>>>>
> >>>>>>>> Can you please re-run tests on your machines using latest
CVS? i
> >>>>>>>> just
> >>>>>>>> fixed http://nagoya.apache.org/jira/browse/AXIS-1202
and that
> >>>>>>>> could be
> >>>>>>>> the root cause of the leaks.
> >>>>>>>>
> >>>>>>>> Thanks,
> >>>>>>>> dims
> >>>>>>>>
> >>>>>>>> On Fri, 16 Jul 2004 22:59:35 -0400, Glen Daniels
> >>>>>>>> <gdaniels@sonicsoftware.com> wrote:
> >>>>>>>>> Me too :)
> >>>>>>>>>
> >>>>>>>>> I just got home and had left an OptimizeIt session
running
> >>>>>>>>> before
> >>>>>>>>> I
> >>>>>>>>> left, so will take a look soon.
> >>>>>>>>>
> >>>>>>>>> --G
> >>>>>>>>>
> >>>>>>>>>
> >>>>>>>>>
> >>>>>>>>>> -----Original Message-----
> >>>>>>>>>> From: Davanum Srinivas [mailto:davanum@gmail.com]
> >>>>>>>>>> Sent: Friday, July 16, 2004 10:58 PM
> >>>>>>>>>> To: Peter Molettiere
> >>>>>>>>>> Cc: Tom Jordahl; dims@apache.org; axis-dev@ws.apache.org;
Glen
> >>>>>>>>>> Daniels
> >>>>>>>>>> Subject: Re: AXIS-1423 : Peter's test case for
performance
> >>>>>>>>>>
> >>>>>>>>>> am still looking :)
> >>>>>>>>>>
> >>>>>>>>>> -- dims
> >>>>>>>>>>
> >>>>>>>>>> On Fri, 16 Jul 2004 18:57:33 -0700, Peter Molettiere
> >>>>>>>>>> <pietro@axonstudios.net> wrote:
> >>>>>>>>>>>
> >>>>>>>>>>> I'll do what I can, although I'm certainly
a bit hampered by
> >>>>>>>>>>> not
> >>>>>>>>>>> having spent lots of time with the axis
source. Someone who
> >>>>>>>>>>> really
> >>>>>>>>>>> understands the axis internals would probably
be able to find
> >>>>>>>>>>> this
> >>>>>>>>>>> much more quickly than I can.
> >>>>>>>>>>>
> >>>>>>>>>>> If you'd like, I'm not far from Macromedia.
You're welcome
> >>>>>>>>>> to come by
> >>>>>>>>>>> and take a look at what I'm doing. Maybe
two heads can come
> >>>>>>>>>> up with a
> >>>>>>>>>>> solution where one can't.
> >>>>>>>>>>>
> >>>>>>>>>>> --Peter
> >>>>>>>>>>>
> >>>>>>>>>>>
> >>>>>>>>>>>
> >>>>>>>>>>> On Jul 16, 2004, at 11:36 AM, Tom Jordahl
wrote:
> >>>>>>>>>>>
> >>>>>>>>>>>>
> >>>>>>>>>>>> Peter,
> >>>>>>>>>>>>
> >>>>>>>>>>>> This is really good work.  It seems
like you really have
> >>>>>>>>>> a handle on
> >>>>>>>>>>>> how to reproduce this problem.
> >>>>>>>>>>>>
> >>>>>>>>>>>> Do you have any tools that you could
bring to bear on
> >>>>>>>>>>>> *where*
> >>>>>>>>>>>> the
> >>>>>>>>>>>> memory leak is in the code?  It sure
sounds like you are
> >>>>>>>>>>>> just a
> >>>>>>>>>>>> single step away from finding the leak.
> >>>>>>>>>>>>
> >>>>>>>>>>>> --
> >>>>>>>>>>>> Tom Jordahl
> >>>>>>>>>>>> Macromedia Server Development
> >>>>>>>>>>>>
> >>>>>>>>>>>> -----Original Message-----
> >>>>>>>>>>>> From: Peter Molettiere [mailto:pietro@axonstudios.net]
> >>>>>>>>>>>> Sent: Friday, July 16, 2004 2:27 PM
> >>>>>>>>>>>> To: dims@apache.org
> >>>>>>>>>>>> Cc: axis-dev@ws.apache.org; Glen Daniels
> >>>>>>>>>>>> Subject: Re: AXIS-1423 : Peter's test
case for performance
> >>>>>>>>>>>>
> >>>>>>>>>>>>
> >>>>>>>>>>>> Hey guys,
> >>>>>>>>>>>>
> >>>>>>>>>>>> We set the numbers high to make the
problem manifest easily,
> >>>>>>>>>>>> however, if you set the return message
size smaller, you'll
> >>>>>>>>>>>> still
> >>>>>>>>>>>> see the server side memory leak, it
will just take much
> >>>>>>>>>>>> longer
> >>>>>>>>>>>> to
> >>>>>>>>>>>> notice, since you have to leak much
more memory before the
> >>>>>>>>>>>> serialization/deserialization overhead
(which is itself
> >>>>>>>>>> much smaller
> >>>>>>>>>>>> due to the smaller message size) bumps
into the max heap
> >>>>>>>>>> size limit.
> >>>>>>>>>>>>
> >>>>>>>>>>>> We're mainly concerned with the server
leaking memory,
> >>>>>>>>>> rather than
> >>>>>>>>>>>> the large amount of memory used during
> >>>>>>>>>>>> serialization/deserialization, since
it eventually crashes
> >>>>>>>>>>>> our
> >>>>>>>>>>>> server, whereas we can just install
more RAM and increase
> >>>>>>>>>> our heap
> >>>>>>>>>>>> size to get around the gobs of memory
used during
> >>>>>>>>>>>> serialization.
> >>>>>>>>>>>>
> >>>>>>>>>>>> --Peter
> >>>>>>>>>>>>
> >>>>>>>>>>>> On Jul 16, 2004, at 9:26 AM, Davanum
Srinivas wrote:
> >>>>>>>>>>>>
> >>>>>>>>>>>>> Glen,
> >>>>>>>>>>>>>
> >>>>>>>>>>>>> Uploaded a slightly revved version
of Peter's test case.
> >>>>>>>>>> Note that
> >>>>>>>>>>>>> there are 15 threads each issuing
50 calls against a
> >>>>>>>>>> service that
> >>>>>>>>>>>>> delivers a 13 MB XML  (Please edit
the client code to
> >>>>>>>>>> restore the
> >>>>>>>>>>>>> original params if needed)
> >>>>>>>>>>>>>
> >>>>>>>>>>>>> http://nagoya.apache.org/jira/browse/AXIS-1423
> >>>>>>>>>>>>>
> >>>>>>>>>>>>> -- dims
> >>>>>>>>>>>>>
> >>>>>>>>>>>>> --
> >>>>>>>>>>>>> Davanum Srinivas - http://webservices.apache.org/~dims/
> >>>>>>>>>>>
> >>>>>>>>>>>
> >>>>>>>>>>
> >>>>>>>>>>
> >>>>>>>>>> --
> >>>>>>>>>> Davanum Srinivas - http://webservices.apache.org/~dims/
> >>>>>>>>>>
> >>>>>>>>>>
> >>>>>>>>>
> >>>>>>>>
> >>>>>>>>
> >>>>>>>> --
> >>>>>>>> Davanum Srinivas - http://webservices.apache.org/~dims/
> >>>>>>>
> >>>>>>>
> >>>>>>
> >>>>>> --
> >>>>>> Davanum Srinivas - http://webservices.apache.org/~dims/
> >>>>>>
> >>>>>
> >>>>>
> >>>>> --
> >>>>> Davanum Srinivas - http://webservices.apache.org/~dims/
> >>>
> >>>
> >>
> >>
> >> -- 
> >> Davanum Srinivas - http://webservices.apache.org/~dims/
>
>

Mime
View raw message