db-derby-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Myrna van Lunteren (JIRA)" <j...@apache.org>
Subject [jira] Commented: (DERBY-4807) analyze performance slow-down of suites.All/derbyall combo between Windows XP and Windows 2003 and Windows 2008.
Date Mon, 06 Dec 2010 20:11:09 GMT

    [ https://issues.apache.org/jira/browse/DERBY-4807?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12968388#action_12968388

Myrna van Lunteren commented on DERBY-4807:

I've not seen trouble with Windows XP...

It's certainly relevant for me to attempt to keep using Windows...

The machine I arranged for has all 3 OSs - Windows XP, Windows 2003, and Windows 2008 installed,
and I caused my nightly run system to kick off a test run from a scheduled task with each
of the OSs loaded. (The OSs were on 1 disk partitioned into 3 equal parts for each of the
OSs; my scripts & thus the derby tests ran on another disk).
I also ran a manual run on each of the OSs, that is, I started RunSuite derbyall, and junit.textui.TestRunner
..suites.All from a shell window.

Some initial findings:
manual 2008                  manual2003                    manualxp
ibm16                               ibm16                               ibm16
derbyall                            derbyall                            derbyall
duration: 00:49:14         00:48:42                           00:51:02
suites.All                          suites.All                          suites.All
duration: 02:05                02:02                                02:12

ibm16                               ibm15                               weme62
derbyall                            derbyall                            derbyall
duration: 02:56:16         02:45:47                           01:24:38
suites.All                          suites.All                          suites.All
duration: 11:03               11:50                                 04:36

ibm16                               ibm15                               weme62
derbyall                            derbyall                            derbyall
duration: 01:51:20         01:42:08                           00:29:48
suites.All                         suites.All                           suites.All
duration: 06:09               05:23                                 00:42

ibm16                               ibm15                               weme62
derbyall		derbyall		derbyall
duration: 00:50:41          00:48:53                           00:30:57
suites.All                          suites.All                          suites.All
duration: 02:12               02:14                                 00:51

Initial conclusions:
With XP, the test runs are indeed quicker when kicked off from a scheduled task - it's not
just an impression. 
This diference is not observed when the tests are kicked off manually.
The one exception is weme62 / 2003 - it's about the same speed as the XP run.

On Windows 2003 and 2008, one can specify 'priority' on a scheduled task (which I did), I
don't know yet exactly what that means, or how this is passed on to underlying processes.
The scheduled task does not directly kick off the java processes, so it's possible something
in the chain of processes doesn't know how to pass on the priority.

I'm continuing analysis & experiments.

> analyze performance slow-down of suites.All/derbyall combo between Windows XP and Windows
2003 and Windows 2008.
> ----------------------------------------------------------------------------------------------------------------
>                 Key: DERBY-4807
>                 URL: https://issues.apache.org/jira/browse/DERBY-4807
>             Project: Derby
>          Issue Type: Task
>          Components: Test
>    Affects Versions:
>         Environment: Windows XP, Windows 2003, Windows 2008, IBM JVM 1.6 (SR8), 1.5 (SR11)
and weme 6.2
>            Reporter: Myrna van Lunteren
>            Assignee: Myrna van Lunteren
> Setting up automated nightly test runs has been troublesome for 10.6 on windows because
the set of tests (derbyall/suitesall with 3 different jvms) that completed within a day on
a machine with similar hardware but running XP did not complete within a day on a machine
running Windows 2008.
> Switching to a machine with windows 2003 appeared to improve matters. Switching to a
machine with XP, and the test finished.
> Initially, my focus has been to try to get the tests to run consistently, now I will
be analyzing this further.
> I now have a separate machine that has the 3 OSs loaded on it, and will experiment, and
gather data, using the 10.6.2 jars so the only difference is the OS.

This message is automatically generated by JIRA.
You can reply to this email to add a comment to the issue online.

View raw message