Re: [BENCHMARK] 2.5.40-mm1 with contest

From: Con Kolivas (conman@kolivas.net)
Date: Wed Dec 04 2002 - 04:28:44 EST


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

On Wed, 4 Dec 2002 07:43 pm, Giuliano Pochini wrote:
> On 03-Dec-2002 Con Kolivas wrote:
> > UP results
> >
> > process_load:
> > Kernel [runs] Time CPU% Loads LCPU% Ratio
> > 2.4.20 [5] 108.1 58 84 40 1.62
> > 2.5.50-mm1 [5] 86.6 78 18 20 1.30
>
> Hm, load task gets half cpu time, but it goes 5 times slower
> in 2.5.x. Why ? You can see a similar behaviour in other
> tests too.

You have to dig deeper to understand why. The time taken to compile the kernel
takes a fixed amount of cpu time. In the presence of a load, it takes longer
in wall clock time, but still takes about the same amount of cpu time. The
amount of extra wall clock time will basically be used for the load,
scheduling, io etc. Now the absolute extra wall clock time it takes to
compile the kernel in this load is time it can also be doing the load.
Therefore, if I spend 20 seconds longer to compile the kernel while the load
is running, the load must also get 20 seconds where it can be doing it's
work. Assuming it does 1 load per second, it will do 20 loads. If it takes 40
seconds longer to compile the kernel, the load gets 40 seconds longer; hence
it can do 40 loads.

Look at the example above and you'll see those numbers. It takes 20 seconds
longer in 2.5.50-mm1 compared to noload, and load gets to do 20 workloads.
2.4.20 takes 40 seconds longer and gets to do 40 workloads. If you take into
account the work done / time they are doing workloads at about the same rate.
Now if one had taken 20 seconds longer than the other and done only the same
amount of work it would be showing serious inefficiencies over and above the
fair scheduling issues which contest is really trying to measure.

Hmm. Not sure if I made that clear enough, but hopefully I got my point
across.

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

iD8DBQE97crOF6dfvkL3i1gRAhdzAKCX8vlHqLQUm+MnzsGAjzP7UPJB4ACbB4um
XyNURkBWQwIC7xAvgkTwmpY=
=4mwU
-----END PGP SIGNATURE-----
-
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/



This archive was generated by hypermail 2b29 : Sat Dec 07 2002 - 22:00:18 EST