[PATCH] V-4.0 Single Priority Array O(1) CPU Scheduler Evaluation

From: Peter Williams
Date: Sun Aug 08 2004 - 23:39:26 EST


Version 4.0 of the various single priority array scheduler patches for 2.6.7 and 2.6.8-rc3 kernels are now available for evaluation.

This version is a fairly major rationalization and simplification of the code with a view to reducing overhead and addressing wli's reservations with respect to the promotion mechanism.

1. [ZAPHOD] My proposed replacement scheduler which offers runtime selectable choice between a priority based or entitlement based O(1) scheduler with active/expired arrays replaced by a single array and an O(1) promotion mechanism plus scheduling statistics with new simplified interactive bonus mechanism and throughput bonus mechanism:

2.6.7
<http://prdownloads.sourceforge.net/cpuse/patch-2.6.7-spa_zaphod_FULL-v4.0?download>
2.6.8-rc3
<http://prdownloads.sourceforge.net/cpuse/patch-2.6.8-rc3-spa_zaphod_FULL-v4.0?download>

The following table of hackbench results for various configurations (IAB column is the maximum allowed interactive bonus and TPB is the maximum allowed throughput bonus) illustrates this scheduler's performance compared to the normal O(1) scheduler.

Kernel Mode IAB TPB 10 groups 100 groups
------------------------------------------------------------
vanilla 1.583(100.0%, 2.2%) 16.740(100.0%, 1.5%)
zaphod "eb" 10 5 1.612(101.8%, 1.0%) 16.422( 98.1%, 1.4%)
zaphod "pb" 10 5 1.570( 99.1%, 0.8%) 16.100( 96.1%, 0.7%)
zaphod "eb" 0 0 1.575( 99.5%, 0.1%) 15.992( 95.5%, 1.0%)
zaphod "pb" 0 0 1.602(101.2%, 5.0%) 15.983( 95.4%, 0.1%)
zaphod "eb" 10 0 1.612(101.8%, 0.1%) 16.142( 96.4%, 0.3%)
zaphod "pb" 10 0 1.584(100.0%, 0.1%) 16.014( 95.6%, 0.01%)
zaphod "eb" 0 5 1.609(101.6%, 0.4%) 16.487( 98.4%, 0.9%)
zaphod "pb" 0 5 1.606(101.4%, 0.5%) 16.367( 97.7%, 1.3%)

The numbers are the results of averaging over 10 runs with the numbers in brackets being the size of the value compared to that for the vanilla kernel and the standard deviation of the value as a percentage. The results for 10 groups show there's no significant difference between the schedulers with the differences being mostly less than half a standard deviation. However, for 100 groups several of the results have differences greater than 1 standard deviation.

Other scheduling parameters were left with their default values but further experimentation will be done to determine their efficacy.

2. Slightly modified version of Con Kolivas's staircase O(1) scheduler
with active/expired arrays replaced by a single array and an O(1)
promotion mechanism:

2.6.7
<http://prdownloads.sourceforge.net/cpuse/patch-2.6.7-spa_sc_FULL-v4.0?download>
2.6.8-rc3
<http://prdownloads.sourceforge.net/cpuse/patch-2.6.8-rc3-spa_sc_FULL-v4.0?download>

3. [HYDRA] Runtime selection between staircase, priority based and
entitlement based O(1) schedulers:

2.6.7
<http://prdownloads.sourceforge.net/cpuse/patch-2.6.7-spa_hydra_FULL-v4.0?download>
2.6.8-rc3
<http://prdownloads.sourceforge.net/cpuse/patch-2.6.8-rc3-spa_hydra_FULL-v4.0?download>

Other schedulers are also available from
<https://sourceforge.net/projects/cpuse/>

So as not to interfere with the staircase scheduler's evaluation I do
not propose to release patches for rc3-mm kernels unless requested.

--
Peter Williams pwil3058@xxxxxxxxxxxxxx

"Learning, n. The kind of ignorance distinguishing the studious."
-- Ambrose Bierce


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/