Skip to main content

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [List Home]
Re: [cross-project-issues-dev] Performance, 3.8 versus 4.2

Another thing to consider is that the Eclipse platform build will using the CBI at some point in the (near?) future.  Is it worth the effort to reconfigure all the performance related tests using PDE/Build if these are going to be done using CBI?

Originally the plan was to have CBI in place for the Juno SR2 builds. If this is still the plan, and performance tests are part of the CBI infrastructure, then maybe we can hold off a few more months.  Of course, if performance tests are not part of CBI, then we certainly should push forward with setting up PDE/Build performance tests on the foundation hardware.

Cheers,
Ian

On Wed, Sep 5, 2012 at 5:48 PM, Denis Roy <denis.roy@xxxxxxxxxxx> wrote:
On 09/05/2012 05:12 PM, Mike Wilson wrote:

2) Working through the issues caused by running the performance tests on shared devices.

[snip]

In a world with potentially other tasks running on the same machines, wildly variable network traffic, etc. I don't think our current performance testing story will work.


Before we discount performance tests on shared hardware, it might be worth a try.

In https://bugs.eclipse.org/33359 I set up a virtual server with one dedicated CPU core and a few gigabytes of dedicated RAM as a proof-of-concept.  The same host server also supports other virtual servers used by hudson.eclipse.org as Hudson slaves.

When I timed a 3.5 second operation executed repeatedly over the course of a normal 24-hour period (24,000 times), the results were remarkable consistent (to a few hundredths of a second).  The actual methodology and results are described in comment https://bugs.eclipse.org/bugs/show_bug.cgi?id=333594#c20.

If need be, we can dedicate a physical disk to the virtual server if startup times or disk I/O need to be benchmarked, and we can dedicate additional CPU cores and RAM if the current setup is insufficient.

Physical hardware dedicated to a single task is costly in terms of maintenance, rack space, power and cooling.

Denis



If that's true, it means it will be a *lot* of work to get them running again. Btw, if anyone has good insights on this and/or wants to help us get the tests running again, we'd love to get your help.

McQ.

Inactive
          hide details for "Andrey Loskutov" ---2012/09/05
          16:16:17---Hi, Listening to all this 4.2 performance
          discussions here"Andrey Loskutov" ---2012/09/05 16:16:17---Hi, Listening to all this 4.2 performance discussions here and for example at

From: "Andrey Loskutov" <loskutov@xxxxxx>
To: cross-project-issues-dev@xxxxxxxxxxx, cross-project-issues-dev-request@xxxxxxxxxxx
Date: 2012/09/05 16:16
Subject: Re: [cross-project-issues-dev] Performance, 3.8 versus 4.2
Sent by: cross-project-issues-dev-bounces@xxxxxxxxxxx





Hi,

Listening to all this 4.2 performance discussions here and for example at
[1] I would like to ask if the is a plan to re-enable performance
regression tests for Eclipse (3.8.x / 4.2.x) platform as we had in the
past before they were disabled in Juno (see [2]).

If there is no such plan yet, shouldn't we have one?

[1]
https://bugs.eclipse.org/bugs/show_bug.cgi?id=385272
[2]
http://wiki.eclipse.org/Platform-releng/Transition_Plans_for_Platform_builds_after_Juno_M6

Regards,
Andrey

On Wed, 05 Sep 2012 15:29:31 +0200,
<cross-project-issues-dev-request@xxxxxxxxxxx> wrote:

> Date: Wed, 5 Sep 2012 09:21:10 -0400
> From: John Arthorne <John_Arthorne@xxxxxxxxxx>
> To: Cross project issues <cross-project-issues-dev@xxxxxxxxxxx>
> Subject: Re: [cross-project-issues-dev] Performance, 3.8 versus 4.2
> Message-ID:
> <OF6B7596FB.B62EF228-ON85257A70.0048E946-85257A70.0049582F@xxxxxxxxxx>
> Content-Type: text/plain; charset="us-ascii"
>
> I suggest anyone having problems to add constructive details on that bug.
> For example profiler output when repeatedly performing a slow operation,
> what plugins are installed, whether it is reproducible with vanilla
> Eclipse SDK, etc. There are some users reporting pervasive slowdowns, and
> for many others it is performing well. Something like a listener leak
> could have effects like this in conjunction with particular installed
> plugins. It takes time after any major release to isolate and resolve
> problems like this.
>
> John
>
> On Wed, Sep 5, 2012 at 3:08 PM, Thomas Hallgren <thomas@xxxxxxx> wrote:
> Hi,
>
> For various reasons I had to switch my development environment from 4.2
> to
> 3.8 today. I was stunned by the performance improvement after the switch.
> The 3.8 platform is much MUCH faster. It boots faster, it closes windows
> faster, it shows menus faster, etc. It also seems to consume less memory
> and be less buggy. The way things stand right now, there's just no way
> I'll switch back to 4.2!
>
> I must say I was very surprised by this. Why is the 4.2 platform what's
> being fronted on the Eclipse download page when it's user experience and
> quality is lagging behind this much? Is it just me who have had this
> experience?
>
> Regards,
> Thomas Hallgren
>
> ------------------------------
>
> Message: 5
> Date: Wed, 5 Sep 2012 06:29:22 -0700
> From: "Konstantin Komissarchik" <konstantin.komissarchik@xxxxxxxxxx>
> To: "'Cross project issues'" <cross-project-issues-dev@xxxxxxxxxxx>
> Subject: Re: [cross-project-issues-dev] Performance, 3.8 versus 4.2
> Message-ID: <001201cd8b6a$76b74950$6425dbf0$@komissarchik@xxxxxxxxxx>
> Content-Type: text/plain; charset="utf-8"
>
> Thomas,
>
>
> You are certainly not the only one seeing performance issues with 4.2. I
> go back and forth between 4.2 and 3.8 every day depending on the project
> I need to work on and the difference is quiet noticeable even on very
> fast hardware. The part I notice the most is the lengthy close all
> editors process. After drilling down into some task and opening a few
> dozen editors, clearing workbench of open editors takes several seconds.
> I can literally watch tabs disappear one by one. The same operation is
> practically instantaneous on 3.8.
>
>
> For stability, user experience and performance reasons, you will find
> that many third party distros have stayed on 3.8 for Juno.
>
>
> I don?t begrudge 4.x its growing pains. It is a complex technological
> shift with a lot of promise. What I find most troubling is the decision
> process that led to the use of 4.2 for Juno distros. When the decision
> was made, it was plainly evident that 4.2 wasn?t going to match 3.8 on
> any of the quality metrics. IDE users might have been ok with quality
> drop if 4.2 delivered compelling new functionality that you couldn?t get
> in 3.8, yet there is no tangible functional delta. The value of 4.x
> platform is for RCP developers and to certain limited extent for IDE
> plugin developers. Certainly not for IDE users. The refreshed
> look-n-feel has been touted as a big end user feature of 4.2, but the
> new look-n-feel itself has numerous issues that leave it looking like an
> unfinished project.
>
>
> Sadly, the user reaction that we?ve been seeing over the last several
> months has been entirely predictable.
>
>
> - Konstantin

--
Kind regards,
Mit freundlichen Grüßen
Andrey Loskutov

+Andrey:
http://plus.google.com/u/0/113794713998126448910
_______________________________________________
cross-project-issues-dev mailing list
cross-project-issues-dev@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/cross-project-issues-dev




_______________________________________________
cross-project-issues-dev mailing list
cross-project-issues-dev@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/cross-project-issues-dev


_______________________________________________
cross-project-issues-dev mailing list
cross-project-issues-dev@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/cross-project-issues-dev




--
R. Ian Bull | EclipseSource Victoria | +1 250 477 7484
http://eclipsesource.com | http://twitter.com/eclipsesource

Back to the top