Re: Local Properties again
On Thu, 3 Jul 2008, Matt Benson [EMAIL PROTECTED] wrote: Surprising, actually, but given that this is the only noise I've seen on the thread I'd say that puts us in good shape to commit. :) +1 Stefan - To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
Re: Local Properties again
--- Dominique Devienne [EMAIL PROTECTED] wrote:
On Wed, Jul 2, 2008 at 3:50 PM, Matt Benson
[EMAIL PROTECTED] wrote:
I would be glad to run the diagnostics if given a
setup or at least your task-level performance
analyzer.
I've uploaded a jar as attachment to bug 23942 w/ my
timer listener. I
don't have a build setup to simulate high property
usage though.
I created setproperties.xml:
project xmlns:ac=antlib:net.sf.antcontrib
ac:for param=n end=
sequential
property name=[EMAIL PROTECTED] value=@{n} /
/sequential
/ac:for
/project
and build.xml:
project
import file=setproperties.xml /
/project
I ran the test three times each on Ant 1.6.2, Kev's
1.7.1 RC, and trunk. The property results are:
162Timings1:306.0 ms 27.4% 1x property
162Timings2:294.0 ms 26.0% 1x property
162Timings3:308.0 ms 27.3% 1x property
171Timings1:278.0 ms 24.9% 1x property
171Timings2:285.0 ms 25.5% 1x property
171Timings3:273.0 ms 24.5% 1x property
trunkTimings1:278.0 ms 22.7% 1x property
trunkTimings2:267.0 ms 21.8% 1x property
trunkTimings3:280.0 ms 22.9% 1x property
So, while I haven't even checked 1.6.3, 1.6.5, or
1.7.0 (the past is the past), it appears that both the
impending release and the trunk outperform 1.6.2.
Surprising, actually, but given that this is the only
noise I've seen on the thread I'd say that puts us in
good shape to commit. :)
-Matt
--DD
-
To unsubscribe, e-mail:
[EMAIL PROTECTED]
For additional commands, e-mail:
[EMAIL PROTECTED]
-
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
Re: Local Properties again
On Thu, Jul 3, 2008 at 11:20 AM, Matt Benson [EMAIL PROTECTED] wrote: So, while I haven't even checked 1.6.3, 1.6.5, or 1.7.0 (the past is the past), it appears that both the impending release and the trunk outperform 1.6.2. Excellent. Thanks for checking. +1. --DD - To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
Local Properties again
Last summer Peter attached an improved patch to bug 23942, which unfortunately still suffered some memory leakage. Using a heap analyzer I was able to, apparently, eliminate these. Dominique reminded me of the performance problems that had been noted by Jan Steve; returning to that thread (1) the last response was that Steve was testing out Peter's fixes. I would like to think no news was good news here; further I recently removed PropertyHelper synchronization that the team had noted was unnecessary. With all this in mind, I would now like to commit a slightly modified version of Peter's patch to trunk. Are there any concerns? -Matt http://markmail.org/message/ivjlvnqmygg4ap5f - To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
Re: Local Properties again
On Wed, Jul 2, 2008 at 2:31 PM, Matt Benson [EMAIL PROTECTED] wrote: Dominique reminded me of the performance problems that had been noted by Jan Steve; ... http://markmail.org/message/ivjlvnqmygg4ap5f Actually it was http://markmail.org/message/rokgze4tfmwrwjab that I had in mind, which was more related to property. As mentioned there, having 150x property in a script imported a number of times by sub-builds (typical of large builds, especially the ones involving native code) could result in 15K invocation of property, taking a large amount of time. So not really related to property access, more property setting. I'd be interested to know whether the patch improves or degrades performance of this particular use case. --DD - To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
Re: Local Properties again
--- Dominique Devienne [EMAIL PROTECTED] wrote: On Wed, Jul 2, 2008 at 2:31 PM, Matt Benson [EMAIL PROTECTED] wrote: Dominique reminded me of the performance problems that had been noted by Jan Steve; ... http://markmail.org/message/ivjlvnqmygg4ap5f Actually it was http://markmail.org/message/rokgze4tfmwrwjab that I had in mind, which was more related to property. As mentioned there, having 150x property in a script imported a number of times by sub-builds (typical of large builds, especially the ones involving native code) could result in 15K invocation of property, taking a large amount of time. So not really related to property access, more property setting. I'd be interested to know whether the patch improves or degrades performance of this particular use case. --DD I would be glad to run the diagnostics if given a setup or at least your task-level performance analyzer. I would suspect that given the improvments I cited, performance should be slightly worse than what you've noted here; however it does seem that with that a good three years in the past it wasn't taken as the emergency the recent case was. -Matt - To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED] - To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
Re: Local Properties again
On Wed, Jul 2, 2008 at 3:50 PM, Matt Benson [EMAIL PROTECTED] wrote: I would be glad to run the diagnostics if given a setup or at least your task-level performance analyzer. I've uploaded a jar as attachment to bug 23942 w/ my timer listener. I don't have a build setup to simulate high property usage though. --DD - To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
