2:44 p.m.
I've been playing with this but wanted to check here first: has anyone
tried running the automated Eclipse tests against the native builds?
I'm referring to the these tests:
http://download.eclipse.org/eclipse/downloads/drops/S-3.1M5a-200502191500...
http://download.eclipse.org/eclipse/downloads/drops/S-3.1M5a-200502191500...
--
Ziga
1:40 a.m.
On Sat, 2005-03-05 at 21:44 +0100, Ziga Mahkovec wrote:
I've been playing with this but wanted to check here first: has
anyone
tried running the automated Eclipse tests against the native builds?
I'm referring to the these tests:
http://download.eclipse.org/eclipse/downloads/drops/S-3.1M5a-200502191500...
http://download.eclipse.org/eclipse/downloads/drops/S-3.1M5a-200502191500...
I'm in the process of finalizing a system to run the automated eclipse
tests on our native builds. Right now it's down to a couple kinks in
the test system and getting some kind of public server set up to show
the results. The system should be ready in the next few days though I
haven't looked into the public server yet (will start looking around
Monday).
Aaron
9:17 a.m.
On Mon, 2005-03-07 at 00:40 -0700, Aaron Luchko wrote:
I'm in the process of finalizing a system to run the automated
eclipse
tests on our native builds. Right now it's down to a couple kinks in
the test system and getting some kind of public server set up to show
the results. The system should be ready in the next few days though I
haven't looked into the public server yet (will start looking around
Monday).
Great, thanks. I ran a subset of these tests and gcj did pretty well.
Might be interesting to run the tests against a non-free JRE as well,
for performance comparison.
--
Ziga
4:20 p.m.
On Mon, 2005-03-07 at 16:17 +0100, Ziga Mahkovec wrote:
On Mon, 2005-03-07 at 00:40 -0700, Aaron Luchko wrote:
> I'm in the process of finalizing a system to run the automated eclipse
> tests on our native builds. Right now it's down to a couple kinks in
> the test system and getting some kind of public server set up to show
> the results. The system should be ready in the next few days though I
> haven't looked into the public server yet (will start looking around
> Monday).
Great, thanks. I ran a subset of these tests and gcj did pretty well.
Might be interesting to run the tests against a non-free JRE as well,
for performance comparison.
I've been pitting them up against the IBM and BEA jvms we have for the
stuff we ship with RHEL-3, the last time I successfully ran them was
with the last 3.0 we had, it didn't fare too badly other than debugging
(no JDWP) though our stuff has changed quite a bit since then.
Aaron
9:15 p.m.
On Mon, Mar 07, 2005 at 03:20:38PM -0700, Aaron Luchko wrote:
I've been pitting them up against the IBM and BEA jvms we have
for the
stuff we ship with RHEL-3, the last time I successfully ran them was
with the last 3.0 we had, it didn't fare too badly other than debugging
(no JDWP) though our stuff has changed quite a bit since then.
"didn't fare too badly" = faster, slower, about the same? Any hard numbers?
--
Dimi.
9:40 a.m.
CCing Bryce McKinlay
* Dimitrie O. Paun <dpaun(a)rogers.com> [2005-03-07 22:16]:
On Mon, Mar 07, 2005 at 03:20:38PM -0700, Aaron Luchko wrote:
> I've been pitting them up against the IBM and BEA jvms we have for the
> stuff we ship with RHEL-3, the last time I successfully ran them was
> with the last 3.0 we had, it didn't fare too badly other than debugging
> (no JDWP) though our stuff has changed quite a bit since then.
"didn't fare too badly" = faster, slower, about the same? Any hard numbers?
I don't have any hard numbers, but things feel about the same or faster
than with a proprietary JVM. The thing to remember is that we have no gcj
optimizations turned on at this point. I know Bryce was working on it ...
Bryce, do you have a status of that?
Andrew
9:50 a.m.
Andrew Overholt wrote:
CCing Bryce McKinlay
* Dimitrie O. Paun <dpaun(a)rogers.com> [2005-03-07 22:16]:
>On Mon, Mar 07, 2005 at 03:20:38PM -0700, Aaron Luchko wrote:
>
>
>>I've been pitting them up against the IBM and BEA jvms we have for the
>>stuff we ship with RHEL-3, the last time I successfully ran them was
>>with the last 3.0 we had, it didn't fare too badly other than debugging
>>(no JDWP) though our stuff has changed quite a bit since then.
>>
>>
>"didn't fare too badly" = faster, slower, about the same? Any hard
numbers?
>
>
I don't have any hard numbers, but things feel about the same or faster
than with a proprietary JVM. The thing to remember is that we have no gcj
optimizations turned on at this point. I know Bryce was working on it ...
Bryce, do you have a status of that?
My last eclipse -O2 attempt was a few weeks ago, and some fixes have
gone in to GCJ since then. At this point it should be possible to
compile most of the .jar's with -O2, but there are likely to still be
optimization bugs that cause eclipse to misbehave or crash when it is
built with -O2. Is resources.jar still used entirely uncompiled? We
should perhaps concentrate on that bug first.
Startup time is also an issue - we should be able to beat the JVM easily
here but currently we arn't. It would be worth profiling the startup
sequence to see if any obvious bottlenecks can be identified.
Bryce
11:55 a.m.
Bryce McKinlay writes:
Startup time is also an issue - we should be able to beat the JVM easily
here but currently we arn't. It would be worth profiling the startup
sequence to see if any obvious bottlenecks can be identified.
The unwinding fixes aren't all in yet. Unwinding accounted for about
16% of startup time in the case of jonas, and I exepct Eclipse would
be similar.
Andrew.
11:56 a.m.
Andrew Haley writes:
Bryce McKinlay writes:
>
> Startup time is also an issue - we should be able to beat the JVM easily
> here but currently we arn't. It would be worth profiling the startup
> sequence to see if any obvious bottlenecks can be identified.
The unwinding fixes aren't all in yet. Unwinding accounted for about
16%
60%
of startup time in the case of jonas, and I exepct Eclipse would
be similar.
Andrew.
12:01 p.m.
* Bryce McKinlay <mckinlay(a)redhat.com> [2005-03-08 10:55]:
My last eclipse -O2 attempt was a few weeks ago, and some fixes have
gone in to GCJ since then. At this point it should be possible to
compile most of the .jar's with -O2, but there are likely to still be
optimization bugs that cause eclipse to misbehave or crash when it is
built with -O2.
I just successfully built all of our jars to .sos with O2 and things ran!
I didn't do much testing, but a simple Hello, world works. Now to see if
this works on the other architectures ...
Is resources.jar still used entirely uncompiled? We should perhaps
concentrate on that bug first.
Yes. Yes, perhaps we should.
Andrew
12:20 p.m.
Bryce> Is resources.jar still used entirely uncompiled? We
Bryce> should perhaps concentrate on that bug first.
I think this bug is still undiagnosed.
Bryce> Startup time is also an issue - we should be able to beat the JVM
Bryce> easily here but currently we arn't. It would be worth profiling the
Bryce> startup sequence to see if any obvious bottlenecks can be identified.
Given Andrew's results from profiling jonas, I would venture to guess
that stack trace performance is a problem. Last time we looked at
Eclipse startup performance, the single biggest improvement was in
this area.
(People sometimes ask why Eclipse startup time matters. The
performance of some other Eclipse operations, like opening a
perspective, is related to the startup time -- a lot of the same sort
of work happens. Improve startup time and you improve some other
visible things in Eclipse.)
Tom
6981
days inactive
6991
days old
java-devel@lists.fedoraproject.org
11 comments
7 participants
participants (7)
-
Aaron Luchko -
Andrew Haley -
Andrew Overholt -
Bryce McKinlay -
Dimitrie O. Paun -
Tom Tromey -
Ziga Mahkovec