Fwd: VM testing with mtest, 2.4.12-ac3, 2.4.12-ac3+riel's patches, and 2.4.13aa1

Fwd: VM testing with mtest, 2.4.12-ac3, 2.4.12-ac3+riel's patches, and 2.4.13aa1

[Josh McKinney]

[-- Attachment #1: Type: text/plain, Size: 1146 bytes --]

This is a report of the mtest01 scripts posted by rwhron@earthlink.net a day orso ago.

Hardware:
K6-3 450
256MB ram
122MB swap
ATA/100 IBM disk

All the tests were run on a fresh boot, in single user mode.  The test consists
of running the mtest script and playing an mpg with mpg123.

The numbers are rather interesting.  While the latency of the ac kernels is
definitely better, the song only dropped out for a second or two in the
begining but that was it.  The aa kernel drops out more frequently throughout
the test, but the amount of memory allocated is almost twice as much as with the
ac kernels.

Hope this report helps, thanks to rwhon for the scripts, seem like a good
testing tool, and thanks to AC for keeping Riel's VM alive, and thanks to Rik
and Andrea for two great VM's.  It is nice to see two competing ideas and how
they fair against each other.

Attached is the vmstat output and mtest summarys for the tests.

--
Linux, the choice                | Pause for storage relocation. 
of a GNU generation       -o)    | 
Kernel 2.4.13-pre3         /\    | 
on a i586                 _\_v   | 
                                 | 

[-- Attachment #2: 2.4.12-ac3.mtest_summary.log --]
[-- Type: text/plain, Size: 329 bytes --]

Averages for 10 mtest01 runs
bytes allocated:                    134427443.2
User time (seconds):                2.546
System time (seconds):              1.370
Elapsed (wall clock) time:          4.798
Percent of CPU this job got:        89.1%
Major (requiring I/O) page faults:  103.8
Minor (reclaiming a frame) faults:  32702

[-- Attachment #3: vmstat-2.4.12-ac3.log.gz --]
[-- Type: application/octet-stream, Size: 2230 bytes --]

[-- Attachment #4: 2.4.12-ac3+riel_patches.mtest_summary.log --]
[-- Type: text/plain, Size: 331 bytes --]

Averages for 10 mtest01 runs
bytes allocated:                    124885401.6
User time (seconds):                2.380
System time (seconds):              1.253
Elapsed (wall clock) time:          4.401
Percent of CPU this job got:        89.1%
Major (requiring I/O) page faults:  100.2
Minor (reclaiming a frame) faults:  30363.3

[-- Attachment #5: vmstat-2.4.12-ac3+riel_patches.log.gz --]
[-- Type: application/octet-stream,, Size: 2322 bytes --]

[-- Attachment #6: 2.4.13-pre3.mtest_summary.log --]
[-- Type: text/plain, Size: 332 bytes --]

Averages for 10 mtest01 runs
bytes allocated:                    288148684.8
User time (seconds):                5.496
System time (seconds):              3.003
Elapsed (wall clock) time:          12.250
Percent of CPU this job got:        68.9%
Major (requiring I/O) page faults:  103.5
Minor (reclaiming a frame) faults:  70380.6

[-- Attachment #7: vmstat-2.4.13-pre3.log.gz --]
[-- Type: application/octet-stream,, Size: 64825 bytes --]

Re: Fwd: VM testing with mtest, 2.4.12-ac3, 2.4.12-ac3+riel's patches, and 2.4.13aa1

[Andreas Dilger]

On Oct 18, 2001  01:51 -0500, Josh McKinney wrote:
> This is a report of the mtest01 scripts posted by rwhron@earthlink.net
> a day orso ago.
> 
> The numbers are rather interesting.  While the latency of the ac kernels is
> definitely better, the song only dropped out for a second or two in the
> begining but that was it.  The aa kernel drops out more frequently throughout
> the test, but the amount of memory allocated is almost twice as much as with
> the ac kernels.

-ac kernel:
> Averages for 10 mtest01 runs
> bytes allocated:                    134427443.2
> User time (seconds):                2.546
> System time (seconds):              1.370
> Elapsed (wall clock) time:          4.798
> Percent of CPU this job got:        89.1%
> Major (requiring I/O) page faults:  103.8
> Minor (reclaiming a frame) faults:  32702

-ac kernel + Rik's patches:
> Averages for 10 mtest01 runs
> bytes allocated:                    124885401.6
> User time (seconds):                2.380
> System time (seconds):              1.253
> Elapsed (wall clock) time:          4.401
> Percent of CPU this job got:        89.1%
> Major (requiring I/O) page faults:  100.2
> Minor (reclaiming a frame) faults:  30363.3

Linus kernel:
> Averages for 10 mtest01 runs
> bytes allocated:                    288148684.8
> User time (seconds):                5.496
> System time (seconds):              3.003
> Elapsed (wall clock) time:          12.250
> Percent of CPU this job got:        68.9%
> Major (requiring I/O) page faults:  103.5
> Minor (reclaiming a frame) faults:  70380.6

Note that the Linus kernel has allocated twice as much memory.  What does that
mean exactly?  The user/system/wall time is also twice as high.  Somehow I
don't think you are having an equal test.

Cheers, Andreas
--
Andreas Dilger  \ "If a man ate a pound of pasta and a pound of antipasto,
                 \  would they cancel out, leaving him still hungry?"
http://www-mddsp.enel.ucalgary.ca/People/adilger/               -- Dogbert

Re: Fwd: VM testing with mtest, 2.4.12-ac3, 2.4.12-ac3+riel's patches, and 2.4.13aa1

[Josh McKinney]

On approximately Thu, Oct 18, 2001 at 09:06:46AM -0600, Andreas Dilger wrote:
> On Oct 18, 2001  01:51 -0500, Josh McKinney wrote:
> > This is a report of the mtest01 scripts posted by rwhron@earthlink.net
> > a day orso ago.
> > 
> > The numbers are rather interesting.  While the latency of the ac kernels is
> > definitely better, the song only dropped out for a second or two in the
> > begining but that was it.  The aa kernel drops out more frequently throughout
> > the test, but the amount of memory allocated is almost twice as much as with
> > the ac kernels.
> 
> -ac kernel:
> > Averages for 10 mtest01 runs
> > bytes allocated:                    134427443.2
> > User time (seconds):                2.546
> > System time (seconds):              1.370
> > Elapsed (wall clock) time:          4.798
> > Percent of CPU this job got:        89.1%
> > Major (requiring I/O) page faults:  103.8
> > Minor (reclaiming a frame) faults:  32702
> 
> -ac kernel + Rik's patches:
> > Averages for 10 mtest01 runs
> > bytes allocated:                    124885401.6
> > User time (seconds):                2.380
> > System time (seconds):              1.253
> > Elapsed (wall clock) time:          4.401
> > Percent of CPU this job got:        89.1%
> > Major (requiring I/O) page faults:  100.2
> > Minor (reclaiming a frame) faults:  30363.3
> 
> Linus kernel:
> > Averages for 10 mtest01 runs
> > bytes allocated:                    288148684.8
> > User time (seconds):                5.496
> > System time (seconds):              3.003
> > Elapsed (wall clock) time:          12.250
> > Percent of CPU this job got:        68.9%
> > Major (requiring I/O) page faults:  103.5
> > Minor (reclaiming a frame) faults:  70380.6
> 
> Note that the Linus kernel has allocated twice as much memory.  What does that
> mean exactly?  The user/system/wall time is also twice as high.  Somehow I
> don't think you are having an equal test.
> 
> Cheers, Andreas

I thought that was strange myself, which is why I metioned it in the begining.
Also when I seen the results I reran the tests, all in single user mode,
freshreboot, put linux single on the CL, voila, sh mtest01.sh & mpg123 some.mp3,
and there is my test.  I know the numbers are crazy and I also know the
importance of having all other variables the same.

Josh
-- 
Linux, the choice                | Animals can be driven crazy by putting too 
of a GNU generation       -o)    | many in too small a pen. Homo sapiens is
Kernel 2.4.13-pre3         /\    | the only animal that voluntarily does this 
on a i586                 _\_v   | to himself.   -- Lazarus Long 
                                 |