[BENCHMARK] scheduler tunables with contest - starvation_limit

[BENCHMARK] scheduler tunables with contest - starvation_limit

[Con Kolivas]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

osdl, contest, tunable - starvation limit on 2.5.52-mm1

noload:
Kernel [runs]           Time    CPU%    Loads   LCPU%   Ratio
sta_lim1000 [3]         39.7    179     0       0       1.10
sta_lim2000 [3]         39.7    181     0       0       1.10

cacherun:
Kernel [runs]           Time    CPU%    Loads   LCPU%   Ratio
sta_lim1000 [3]         36.7    194     0       0       1.01
sta_lim2000 [3]         37.0    194     0       0       1.02

process_load:
Kernel [runs]           Time    CPU%    Loads   LCPU%   Ratio
sta_lim1000 [3]         48.2    146     10      47      1.33
sta_lim2000 [3]         45.6    157     7       37      1.26

ctar_load:
Kernel [runs]           Time    CPU%    Loads   LCPU%   Ratio
sta_lim1000 [3]         57.9    158     1       10      1.60
sta_lim2000 [3]         54.3    153     1       10      1.50

xtar_load:
Kernel [runs]           Time    CPU%    Loads   LCPU%   Ratio
sta_lim1000 [3]         72.9    125     1       8       2.01
sta_lim2000 [3]         67.6    130     1       9       1.87

io_load:
Kernel [runs]           Time    CPU%    Loads   LCPU%   Ratio
sta_lim1000 [3]         89.6    98      12      19      2.47
sta_lim2000 [3]         79.9    104     11      19      2.21

io_other:
Kernel [runs]           Time    CPU%    Loads   LCPU%   Ratio
sta_lim1000 [3]         68.1    124     8       19      1.88
sta_lim2000 [3]         68.5    116     9       20      1.89

read_load:
Kernel [runs]           Time    CPU%    Loads   LCPU%   Ratio
sta_lim1000 [3]         49.9    151     5       6       1.38
sta_lim2000 [3]         50.9    150     5       6       1.41

list_load:
Kernel [runs]           Time    CPU%    Loads   LCPU%   Ratio
sta_lim1000 [3]         43.8    167     0       9       1.21
sta_lim2000 [3]         43.5    168     0       9       1.20

mem_load:
Kernel [runs]           Time    CPU%    Loads   LCPU%   Ratio
sta_lim1000 [3]         106.8   77      36      2       2.95
sta_lim2000 [3]         112.4   73      36      2       3.10

Slight balance changes here. Most io things take slightly longer with lower 
starvation limit and mem_load takes less time.

Con
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.0 (GNU/Linux)

iD8DBQE+AlriF6dfvkL3i1gRAugYAJ93cYDhjqXjM4TIZsLF+zvUtMoJ5QCfS5EC
nIPWPR1JF0awLBCvL1uBzJ4=
=eU+1
-----END PGP SIGNATURE-----

Re: [BENCHMARK] scheduler tunables with contest - starvation_limit

[Robert Love]

On Thu, 2002-12-19 at 18:48, Con Kolivas wrote:

> osdl, contest, tunable - starvation limit on 2.5.52-mm1

Con, curiously, what is this OSDL hardware like?

One thing I always liked about your Contest runs were you did them on
your home machine, which was presumably fairly run-of-the-mill so we
could keep an eye on the low-end desktop machines.

	Robert Love 

Re: [BENCHMARK] scheduler tunables with contest - starvation_limit

[David Lang]

one other thing that would be interesting to test on the osdl machines
would be the effect of different filesystems.

the origional set of tests were all done on reiserfs, it would be
interesting to see if there is a difference between it and the others.

David Lang

On 22 Dec 2002, Robert Love wrote:

> Date: 22 Dec 2002 20:06:51 -0500
> From: Robert Love <rml@tech9.net>
> To: Con Kolivas <conman@kolivas.net>
> Cc: linux kernel mailing list <linux-kernel@vger.kernel.org>
> Subject: Re: [BENCHMARK] scheduler tunables with contest -
>     starvation_limit
>
> On Thu, 2002-12-19 at 18:48, Con Kolivas wrote:
>
> > osdl, contest, tunable - starvation limit on 2.5.52-mm1
>
> Con, curiously, what is this OSDL hardware like?
>
> One thing I always liked about your Contest runs were you did them on
> your home machine, which was presumably fairly run-of-the-mill so we
> could keep an eye on the low-end desktop machines.
>
> 	Robert Love
>
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/
>

Re: [BENCHMARK] scheduler tunables with contest - starvation_limit

[Con Kolivas]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1


>one other thing that would be interesting to test on the osdl machines
>would be the effect of different filesystems.
>
>the origional set of tests were all done on reiserfs, it would be
>interesting to see if there is a difference between it and the others.

The current osdl hardware uses ext3 in the default journalling mode. Trying 
different filesystems is something I have had planned for a while. When I get 
the hardware sorted out as I need it to do this I will post some results 
where comparisons can be made.

See the current specs here:
http://www.osdl.org/projects/ctdevel/results/

Con
>
>David Lang
>
>On 22 Dec 2002, Robert Love wrote:
>> Date: 22 Dec 2002 20:06:51 -0500
>> From: Robert Love <rml@tech9.net>
>> To: Con Kolivas <conman@kolivas.net>
>> Cc: linux kernel mailing list <linux-kernel@vger.kernel.org>
>> Subject: Re: [BENCHMARK] scheduler tunables with contest -
>>     starvation_limit
>>
>> On Thu, 2002-12-19 at 18:48, Con Kolivas wrote:
>> > osdl, contest, tunable - starvation limit on 2.5.52-mm1
>>
>> Con, curiously, what is this OSDL hardware like?
>>
>> One thing I always liked about your Contest runs were you did them on
>> your home machine, which was presumably fairly run-of-the-mill so we
>> could keep an eye on the low-end desktop machines.
>>
>> 	Robert Love
>>
>> -
>> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
>> the body of a message to majordomo@vger.kernel.org
>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>> Please read the FAQ at  http://www.tux.org/lkml/
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.0 (GNU/Linux)

iD8DBQE+BmmrF6dfvkL3i1gRApCRAJ96mtrpTCap5JoCQAX6UB3O39y3bgCcDoXM
XMa0Pz9Ldrdir9LQ4Qj83pI=
=CmgE
-----END PGP SIGNATURE-----

Re: [BENCHMARK] scheduler tunables with contest - starvation_limit

[Robert Love]

On Sun, 2002-12-22 at 20:40, Con Kolivas wrote:

> The current osdl hardware uses ext3 in the default journalling mode. Trying 
> different filesystems is something I have had planned for a while. When I get 
> the hardware sorted out as I need it to do this I will post some results 
> where comparisons can be made.

One thing I have found in doing low-latency research is the impact of
ext3 over ext2.  There is a periodic blip of high latency with ext3 not
seen in ext2.  Presumably due to the journal writeback of ext3.

It was not huge but a measurable increase.  This was on 2.4, so I am
curious how improved it is on 2.5.

	Robert Love

Re: [BENCHMARK] scheduler tunables with contest - starvation_limit

[Con Kolivas]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1


>On Thu, 2002-12-19 at 18:48, Con Kolivas wrote:
>> osdl, contest, tunable - starvation limit on 2.5.52-mm1
>
>Con, curiously, what is this OSDL hardware like?
>
>One thing I always liked about your Contest runs were you did them on
>your home machine, which was presumably fairly run-of-the-mill so we
>could keep an eye on the low-end desktop machines.

Sure. The osdl hardware is better than the laptop I used to use but in
uniprocessor mode is fairly run of the mill. Its a PIII 866x2. The full
hardware specs and archived major version results can be found here to be
perused at your leisure:

http://www.osdl.org/projects/ctdevel/results/

Cheers,
Con
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.0 (GNU/Linux)

iD8DBQE+BmkmF6dfvkL3i1gRAnaPAJ9G3MU/CZQa8ayc7nNy4l7E2y2a4ACfTyCA
XtneOxQtQY9v3N6uKjLa5Hw=
=/REh
-----END PGP SIGNATURE-----

Re: [BENCHMARK] scheduler tunables with contest - starvation_limit

[Con Kolivas]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1


>On Thu, 2002-12-19 at 18:48, Con Kolivas wrote:
>> osdl, contest, tunable - starvation limit on 2.5.52-mm1
>
>Con, curiously, what is this OSDL hardware like?
>
>One thing I always liked about your Contest runs were you did them on
>your home machine, which was presumably fairly run-of-the-mill so we
>could keep an eye on the low-end desktop machines.

Forgot to mention I've been doing the scheduler tunables in smp mode just so 
it wouldnt take me too long to get results. I have no doubt the signal to 
noise ratio is greater in the uniprocessor results though.

Con
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.0 (GNU/Linux)

iD8DBQE+BmtyF6dfvkL3i1gRAq6wAJ9M1DlPK8JL5RaEbaOOcA6z+KhmZwCcDn+W
MUDPF6uqNGyQ8FRtWyr07B4=
=6oDv
-----END PGP SIGNATURE-----