You are viewing a plain text version of this content. The canonical link for it is here.
Posted to java-dev@axis.apache.org by Davanum Srinivas <da...@gmail.com> on 2004/10/28 22:46:15 UTC
Axis 1.2 RC2? Final?
Glen,
Can you please cut a RC2 next week? Then me and Ias can start again on
the TCK testing and we can release 1.2 as soon as we finish with the
TCK. We can shoot for a 1.2 Release before/during ApacheCon (Nov
15-17) if all goes well.
Tom,
Do you want to take a crack at exposing the jaxrpc compliance flag via
the tools?
Folks,
If you have patches that have not yet been committed. Please let me
know ASAP by adding a comment in JIRA. (After making sure that the
patches can be applied to current CVS)
Thanks,
dims
--
Davanum Srinivas - http://webservices.apache.org/~dims/
Re: Axis 1.2 RC2? Final?
Posted by Davanum Srinivas <da...@gmail.com>.
Glen,
I just uploaded a zip with a stand-alone test case that can be run
against both 1.1 and 1.2 using SimpleAxisServer and dynamically
generated stubs (for http://issues.apache.org/jira/browse/AXIS-1608).
Bottom Line: There is NO problem with performance (at least NOT with
this test case).
Thanks,
-- dims
On Tue, 2 Nov 2004 10:06:17 -0800, Glen Daniels <gl...@thoughtcraft.com> wrote:
>
> This sounds like a good plan to me, dims, except:
>
> My main worry at this point is the performance degradation we're seeing in
> 1.2 (http://issues.apache.org/jira/browse/AXIS-1608). I don't feel
> comfortable releasing 1.2 final until we've resolved this. Is anyone
> focusing on this issue right now? I can try to look at it this week, but
> I'm not on the greatest performance-analysis machine (I'm still travelling).
> Just getting a couple of good profiling runs with Axis 1.1 and then 1.2
> would be really useful, I bet.
>
> Anyway, I'll plan to release 1.2RC2 this week, probably Friday. If we
> manage to nail the performance problem, perhaps RC2 will be good enough to
> be 1.2Final. Otherwise, we might need RC3 then Final.
>
> Does this seem reasonable? Everyone else agree that the performance issue
> is really major?
>
> --Glen
>
>
>
> > -----Original Message-----
> > From: Davanum Srinivas [mailto:davanum@gmail.com]
> > Sent: Thursday, October 28, 2004 1:46 PM
> > To: axis-dev@ws.apache.org
> > Subject: Axis 1.2 RC2? Final?
> >
> > Glen,
> > Can you please cut a RC2 next week? Then me and Ias can start
> > again on the TCK testing and we can release 1.2 as soon as we
> > finish with the TCK. We can shoot for a 1.2 Release
> > before/during ApacheCon (Nov
> > 15-17) if all goes well.
> >
> > Tom,
> > Do you want to take a crack at exposing the jaxrpc compliance
> > flag via the tools?
> >
> > Folks,
> > If you have patches that have not yet been committed. Please
> > let me know ASAP by adding a comment in JIRA. (After making
> > sure that the patches can be applied to current CVS)
> >
> > Thanks,
> > dims
> >
> > --
> > Davanum Srinivas - http://webservices.apache.org/~dims/
> >
> >
>
--
Davanum Srinivas - http://webservices.apache.org/~dims/
Re: Axis 1.2 RC2? Final?
Posted by Davanum Srinivas <da...@gmail.com>.
Ok. i did some profiling and committed a change
(http://nagoya.apache.org/eyebrowse/ReadMsg?listName=axis-cvs@ws.apache.org&msgNo=3275).
let me know if there's any change in the numbers. Also, am finding
that for this case JDK13 has the best performance and JDK15 is the
worst...So make sure you run your tests always against the same JDK to
compare numbers.
thanks,
dims
On Wed, 3 Nov 2004 08:55:18 -0500, Davanum Srinivas <da...@gmail.com> wrote:
> try to modify the test.encoding.TestString2 with your 1 MB message and
> use a profiler to figure out what's taking the CPU time.
>
> thanks,
> dims
>
>
>
>
> On Wed, 3 Nov 2004 18:27:46 +0530, Venkat Reddy <vr...@gmail.com> wrote:
> > One of the major problem areas seems to be the DeserializationContext
> > class. In Axis 1.1 DeserializationContextImpl appears to work more
> > efficiently. This observation is based on the following numbers:
> >
> > The time elapsed between entering AxisServlet.doPost() and Entering
> > Java.Provider.Invoke() is as follows:
> >
> > Axis 1.1: 62 ms
> > Axis 1.2: 219 ms
> >
> > That's the time spent for bulk of deserialization invocations. This
> > difference is observed with 1 MB message size, and is almost nullifed
> > for very small message sizes (upto 50k).
> >
> > -- Venkat
> >
> >
> >
> >
> > On Tue, 2 Nov 2004 10:06:17 -0800, Glen Daniels <gl...@thoughtcraft.com> wrote:
> > >
> > > This sounds like a good plan to me, dims, except:
> > >
> > > My main worry at this point is the performance degradation we're seeing in
> > > 1.2 (http://issues.apache.org/jira/browse/AXIS-1608). I don't feel
> > > comfortable releasing 1.2 final until we've resolved this. Is anyone
> > > focusing on this issue right now? I can try to look at it this week, but
> > > I'm not on the greatest performance-analysis machine (I'm still travelling).
> > > Just getting a couple of good profiling runs with Axis 1.1 and then 1.2
> > > would be really useful, I bet.
> > >
> > > Anyway, I'll plan to release 1.2RC2 this week, probably Friday. If we
> > > manage to nail the performance problem, perhaps RC2 will be good enough to
> > > be 1.2Final. Otherwise, we might need RC3 then Final.
> > >
> > > Does this seem reasonable? Everyone else agree that the performance issue
> > > is really major?
> > >
> > > --Glen
> > >
> > >
> > >
> > > > -----Original Message-----
> > > > From: Davanum Srinivas [mailto:davanum@gmail.com]
> > > > Sent: Thursday, October 28, 2004 1:46 PM
> > > > To: axis-dev@ws.apache.org
> > > > Subject: Axis 1.2 RC2? Final?
> > > >
> > > > Glen,
> > > > Can you please cut a RC2 next week? Then me and Ias can start
> > > > again on the TCK testing and we can release 1.2 as soon as we
> > > > finish with the TCK. We can shoot for a 1.2 Release
> > > > before/during ApacheCon (Nov
> > > > 15-17) if all goes well.
> > > >
> > > > Tom,
> > > > Do you want to take a crack at exposing the jaxrpc compliance
> > > > flag via the tools?
> > > >
> > > > Folks,
> > > > If you have patches that have not yet been committed. Please
> > > > let me know ASAP by adding a comment in JIRA. (After making
> > > > sure that the patches can be applied to current CVS)
> > > >
> > > > Thanks,
> > > > dims
> > > >
> > > > --
> > > > Davanum Srinivas - http://webservices.apache.org/~dims/
> > > >
> > > >
> > >
> >
>
>
> --
>
>
> Davanum Srinivas - http://webservices.apache.org/~dims/
>
--
Davanum Srinivas - http://webservices.apache.org/~dims/
Re: Axis 1.2 RC2? Final?
Posted by Davanum Srinivas <da...@gmail.com>.
try to modify the test.encoding.TestString2 with your 1 MB message and
use a profiler to figure out what's taking the CPU time.
thanks,
dims
On Wed, 3 Nov 2004 18:27:46 +0530, Venkat Reddy <vr...@gmail.com> wrote:
> One of the major problem areas seems to be the DeserializationContext
> class. In Axis 1.1 DeserializationContextImpl appears to work more
> efficiently. This observation is based on the following numbers:
>
> The time elapsed between entering AxisServlet.doPost() and Entering
> Java.Provider.Invoke() is as follows:
>
> Axis 1.1: 62 ms
> Axis 1.2: 219 ms
>
> That's the time spent for bulk of deserialization invocations. This
> difference is observed with 1 MB message size, and is almost nullifed
> for very small message sizes (upto 50k).
>
> -- Venkat
>
>
>
>
> On Tue, 2 Nov 2004 10:06:17 -0800, Glen Daniels <gl...@thoughtcraft.com> wrote:
> >
> > This sounds like a good plan to me, dims, except:
> >
> > My main worry at this point is the performance degradation we're seeing in
> > 1.2 (http://issues.apache.org/jira/browse/AXIS-1608). I don't feel
> > comfortable releasing 1.2 final until we've resolved this. Is anyone
> > focusing on this issue right now? I can try to look at it this week, but
> > I'm not on the greatest performance-analysis machine (I'm still travelling).
> > Just getting a couple of good profiling runs with Axis 1.1 and then 1.2
> > would be really useful, I bet.
> >
> > Anyway, I'll plan to release 1.2RC2 this week, probably Friday. If we
> > manage to nail the performance problem, perhaps RC2 will be good enough to
> > be 1.2Final. Otherwise, we might need RC3 then Final.
> >
> > Does this seem reasonable? Everyone else agree that the performance issue
> > is really major?
> >
> > --Glen
> >
> >
> >
> > > -----Original Message-----
> > > From: Davanum Srinivas [mailto:davanum@gmail.com]
> > > Sent: Thursday, October 28, 2004 1:46 PM
> > > To: axis-dev@ws.apache.org
> > > Subject: Axis 1.2 RC2? Final?
> > >
> > > Glen,
> > > Can you please cut a RC2 next week? Then me and Ias can start
> > > again on the TCK testing and we can release 1.2 as soon as we
> > > finish with the TCK. We can shoot for a 1.2 Release
> > > before/during ApacheCon (Nov
> > > 15-17) if all goes well.
> > >
> > > Tom,
> > > Do you want to take a crack at exposing the jaxrpc compliance
> > > flag via the tools?
> > >
> > > Folks,
> > > If you have patches that have not yet been committed. Please
> > > let me know ASAP by adding a comment in JIRA. (After making
> > > sure that the patches can be applied to current CVS)
> > >
> > > Thanks,
> > > dims
> > >
> > > --
> > > Davanum Srinivas - http://webservices.apache.org/~dims/
> > >
> > >
> >
>
--
Davanum Srinivas - http://webservices.apache.org/~dims/
Re: Axis 1.2 RC2? Final?
Posted by Venkat Reddy <vr...@gmail.com>.
One of the major problem areas seems to be the DeserializationContext
class. In Axis 1.1 DeserializationContextImpl appears to work more
efficiently. This observation is based on the following numbers:
The time elapsed between entering AxisServlet.doPost() and Entering
Java.Provider.Invoke() is as follows:
Axis 1.1: 62 ms
Axis 1.2: 219 ms
That's the time spent for bulk of deserialization invocations. This
difference is observed with 1 MB message size, and is almost nullifed
for very small message sizes (upto 50k).
-- Venkat
On Tue, 2 Nov 2004 10:06:17 -0800, Glen Daniels <gl...@thoughtcraft.com> wrote:
>
> This sounds like a good plan to me, dims, except:
>
> My main worry at this point is the performance degradation we're seeing in
> 1.2 (http://issues.apache.org/jira/browse/AXIS-1608). I don't feel
> comfortable releasing 1.2 final until we've resolved this. Is anyone
> focusing on this issue right now? I can try to look at it this week, but
> I'm not on the greatest performance-analysis machine (I'm still travelling).
> Just getting a couple of good profiling runs with Axis 1.1 and then 1.2
> would be really useful, I bet.
>
> Anyway, I'll plan to release 1.2RC2 this week, probably Friday. If we
> manage to nail the performance problem, perhaps RC2 will be good enough to
> be 1.2Final. Otherwise, we might need RC3 then Final.
>
> Does this seem reasonable? Everyone else agree that the performance issue
> is really major?
>
> --Glen
>
>
>
> > -----Original Message-----
> > From: Davanum Srinivas [mailto:davanum@gmail.com]
> > Sent: Thursday, October 28, 2004 1:46 PM
> > To: axis-dev@ws.apache.org
> > Subject: Axis 1.2 RC2? Final?
> >
> > Glen,
> > Can you please cut a RC2 next week? Then me and Ias can start
> > again on the TCK testing and we can release 1.2 as soon as we
> > finish with the TCK. We can shoot for a 1.2 Release
> > before/during ApacheCon (Nov
> > 15-17) if all goes well.
> >
> > Tom,
> > Do you want to take a crack at exposing the jaxrpc compliance
> > flag via the tools?
> >
> > Folks,
> > If you have patches that have not yet been committed. Please
> > let me know ASAP by adding a comment in JIRA. (After making
> > sure that the patches can be applied to current CVS)
> >
> > Thanks,
> > dims
> >
> > --
> > Davanum Srinivas - http://webservices.apache.org/~dims/
> >
> >
>
Re: Axis 1.2 RC2? Final?
Posted by Archit Shah <ar...@alum.mit.edu>.
I've been looking at Axis 1.2 performance for a couple of days. Here
are quick profiling results using java 1.4.2 built-in
hprof:cpu=samples. I tested the 1.1 client with the 1.1 server and the
1.2 client with 1.2 server. These do not qualify as "good profiling
runs". I was using SimpleAxisServer instead of tomcat, not sure if that
is what you are looking for. (also I modified the test to use 1MB
requests instead of 8MB. that didn't affect the relative performance of
1.1 and 1.2).
server 1.2
rank self accum count trace method
1 61.97% 61.97% 3863 239 java.net.PlainSocketImpl.socketAccept
2 15.35% 77.32% 957 672 java.net.PlainSocketImpl.socketClose0
3 6.24% 83.56% 389 671
java.net.SocketOutputStream.socketWrite0
4 1.44% 85.00% 90 30 java.lang.ClassLoader.defineClass0
5 0.85% 85.85% 53 669
org.apache.axis.components.encoding.UTF8Encoder.writeEncoded
server 1.1
rank self accum count trace method
1 65.17% 65.17% 5071 230 java.net.PlainSocketImpl.socketAccept
2 11.59% 76.76% 902 566 java.net.PlainSocketImpl.socketClose0
3 10.02% 86.79% 780 567 java.net.SocketInputStream.socketRead0
4 1.32% 88.11% 103 542 java.net.SocketInputStream.socketRead0
5 1.02% 89.13% 79 23 java.lang.ClassLoader.defineClass0
client 1.2
rank self accum count trace method
1 48.72% 48.72% 1639 626 java.net.SocketInputStream.socketRead0
2 14.89% 63.61% 501 625
java.net.SocketOutputStream.socketWrite0
3 5.83% 69.44% 196 622 java.net.PlainSocketImpl.socketConnect
4 2.91% 72.35% 98 16 java.lang.ClassLoader.defineClass0
5 1.46% 73.81% 49 607
org.apache.axis.components.encoding.UTF8Encoder.writeEncoded
6 1.04% 74.85% 35 55 java.util.zip.Inflater.inflateBytes
7 0.65% 75.51% 22 49
java.lang.ClassLoader.findBootstrapClass
8 0.65% 76.16% 22 649 java.net.SocketInputStream.socketRead0
9 0.62% 76.78% 21 668
sun.nio.cs.UTF_8$Encoder.encodeArrayLoop
10 0.56% 77.35% 19 648 java.io.CharArrayWriter.write
client 1.1
rank self accum count trace method
1 68.92% 68.92% 2599 596 java.net.SocketInputStream.socketRead0
2 2.33% 71.25% 88 15 java.lang.ClassLoader.defineClass0
3 2.33% 73.59% 88 615
org.apache.crimson.parser.XmlReader$Utf8Reader.read
4 1.38% 74.97% 52 595
java.net.SocketOutputStream.socketWrite0
5 1.22% 76.19% 46 612 java.net.SocketInputStream.socketRead0
6 1.03% 77.22% 39 614
org.apache.crimson.parser.InputEntity.parsedContent
7 0.85% 78.07% 32 626 java.io.CharArrayWriter.write
8 0.69% 78.76% 26 26 java.util.zip.Inflater.inflateBytes
9 0.61% 79.37% 23 624 java.lang.System.arraycopy
10 0.50% 79.87% 19 36
java.lang.ClassLoader.findBootstrapClass
The 1.1 server and client spent a lot less time writing. I have no idea
why that is. I imagine issue Axis-1567 is important to the relative
slowness of 1.2 ("the memory consumption goes up by 9MB for evey 1MB of
string size."). I am continuing to look into this issue. If there are
particular numbers that are needed let me know.
-- Archit Shah
On Tuesday, Nov 2, 2004, at 13:06 US/Eastern, Glen Daniels wrote:
> My main worry at this point is the performance degradation we're
> seeing in
> 1.2 (http://issues.apache.org/jira/browse/AXIS-1608). I don't feel
> comfortable releasing 1.2 final until we've resolved this. Is anyone
> focusing on this issue right now? I can try to look at it this week,
> but
> I'm not on the greatest performance-analysis machine (I'm still
> travelling).
> Just getting a couple of good profiling runs with Axis 1.1 and then 1.2
> would be really useful, I bet.
RE: Axis 1.2 RC2? Final?
Posted by Glen Daniels <gl...@thoughtcraft.com>.
This sounds like a good plan to me, dims, except:
My main worry at this point is the performance degradation we're seeing in
1.2 (http://issues.apache.org/jira/browse/AXIS-1608). I don't feel
comfortable releasing 1.2 final until we've resolved this. Is anyone
focusing on this issue right now? I can try to look at it this week, but
I'm not on the greatest performance-analysis machine (I'm still travelling).
Just getting a couple of good profiling runs with Axis 1.1 and then 1.2
would be really useful, I bet.
Anyway, I'll plan to release 1.2RC2 this week, probably Friday. If we
manage to nail the performance problem, perhaps RC2 will be good enough to
be 1.2Final. Otherwise, we might need RC3 then Final.
Does this seem reasonable? Everyone else agree that the performance issue
is really major?
--Glen
> -----Original Message-----
> From: Davanum Srinivas [mailto:davanum@gmail.com]
> Sent: Thursday, October 28, 2004 1:46 PM
> To: axis-dev@ws.apache.org
> Subject: Axis 1.2 RC2? Final?
>
> Glen,
> Can you please cut a RC2 next week? Then me and Ias can start
> again on the TCK testing and we can release 1.2 as soon as we
> finish with the TCK. We can shoot for a 1.2 Release
> before/during ApacheCon (Nov
> 15-17) if all goes well.
>
> Tom,
> Do you want to take a crack at exposing the jaxrpc compliance
> flag via the tools?
>
> Folks,
> If you have patches that have not yet been committed. Please
> let me know ASAP by adding a comment in JIRA. (After making
> sure that the patches can be applied to current CVS)
>
> Thanks,
> dims
>
> --
> Davanum Srinivas - http://webservices.apache.org/~dims/
>
>