Skip to contentSkip to conversation

2.6.17-ck1

DISCUSSION Con Kolivas 3 participants 7 messages Jun 18, 2006 – Jun 19, 2006

lore
Share & tools
lore
#0 lore raw

These are patches designed to improve system responsiveness and interactivity. It is configurable to any workload but the default ck patch is aimed at the desktop and cks is available with more emphasis on serverspace.

Apply to 2.6.17
http://www.kernel.org/pub/linux/kernel/people/ck/patches/2.6/2.6.17/2.6.17-ck1/patch-2.6.17-ck1.bz2

or server version
http://www.kernel.org/pub/linux/kernel/people/ck/patches/cks/patch-2.6.17-cks1.bz2

web:
http://kernel.kolivas.org

all patches:
http://www.kernel.org/pub/linux/kernel/people/ck/patches/

Split patches available.

Full patchlist:

sched-implement-smpnice-2.6.17.patch
sched-revise_smt_nice_locking.patch
2.6.17-smpnice-staircase-16.patch
sched-staircase16_interactive_tunable.patch
sched-staircase16_compute_tunable.patch
sched-range.patch
sched-iso-4.5.patch
track_mutexes-1.patch
sched-idleprio-1.9.patch
sched-limit_policy_changes.patch
defaultcfq.diff
cfq-ioprio_inherit_rt_class.patch
cfq-iso_idleprio_ionice.patch
hz-default_1000.patch
hz-no_default_250.patch
sched-add-above-background-load-function.patch
mm-swap_prefetch-32.patch
swsusp-rework-memory-shrinker-rev-2.patch
mm-convert_swappiness_to_mapped.patch
mm-lots_watermark.diff
mm-kswapd_inherit_prio-1.patch
mm-prio_dependant_scan-1.patch
mm-background_scan-1.patch
mm-idleprio_prio.patch
mm-decrease_minimum_dirty_ratio.patch
mm-set_zero_dirty_ratio.patch
mm-filesize_dependant_lru_cache_add.patch
fs-fcache-v2.1.patch
kconfig-expose_vmsplit_option.patch
2.6.17-ck1-version.patch

Signature · 1 line
-ck
#1 lore raw
Quoted 5 lines

These are patches designed to improve system responsiveness and interactivity. It is configurable to any workload but the default ck patch is aimed at the desktop and cks is available with more emphasis on serverspace.

Last time (2.6.17-rc6-ck1), I had a strange experience over the regular scheduler. When three gccs were competing for time, each of them got a time window of 1/3 second in which each ran at 99%.
The regular scheduler does it in a way so that each time window is as small as possible, that is, top shows 33% for each process on low top udating intervals like 0.1 sec.
This behavior was not observed with 2.6.16-rcX-ck.

Jan Engelhardt

#2 lore raw
Quoted 5 lines · Jan Engelhardt
Quoted 4 lines

These are patches designed to improve system responsiveness and interactivity. It is configurable to any workload but the default ck patch is aimed at the desktop and cks is available with more emphasis on serverspace.

Quoted 7 lines

Last time (2.6.17-rc6-ck1), I had a strange experience over the regular scheduler. When three gccs were competing for time, each of them got a time window of 1/3 second in which each ran at 99%.
The regular scheduler does it in a way so that each time window is as small as possible, that is, top shows 33% for each process on low top udating intervals like 0.1 sec.
This behavior was not observed with 2.6.16-rcX-ck.

Were you running them SCHED_IDLEPRIO or in compute mode? They would do that.

Signature · 1 line
-ck
#3 lore raw
Quoted 3 lines

Were you running them SCHED_IDLEPRIO or in compute mode? They would do that.

I have not changed anything, so I presume SCHED_NORMAL. Unless they have been made SCHED_IDLEPRIO/compute by staircase's logic...

Jan Engelhardt

#4 lore raw
Quoted 5 lines · Jan Engelhardt
Quoted 2 lines

Were you running them SCHED_IDLEPRIO or in compute mode? They would do that.

I have not changed anything, so I presume SCHED_NORMAL. Unless they have been made SCHED_IDLEPRIO/compute by staircase's logic...

No it wouldn't do that.

I've not seen what you describe happening but definitely the timing of parallelising jobs in 'make' completely changes with cpu scheduler changes and with I/O scheduler changes.

However if it's purely unrelated cpu bound tasks running and no disk I/O involved then full timeslices run out at ~114ms at 1000HZ (longer at lower HZ) so that should be the longest period one task of the same priority could possibly run before the others do. That sort of timeslice you would not pick up at .1s interval 'top's but I find 'top' can be very deceiving if its timing happens to be exactly what the intervals are it can look like only one thing is running or one thing is stuck at the same priority and so on.

With actual numbers from interbench testing of fully cpu bound tasks (under what's called benchmarking cpu of Gaming) the average and max scheduling delays look of the same magnitude as mainline.

Signature · 1 line
-ck
#5 lore raw
In reply to Con Kolivas
Quoted 11 lines · Con Kolivas

These are patches designed to improve system responsiveness and interactivity. It is configurable to any workload but the default ck patch is aimed at the desktop and cks is available with more emphasis on serverspace.

Apply to 2.6.17
http://www.kernel.org/pub/linux/kernel/people/ck/patches/2.6/2.6.17/2.6.17-ck1/patch-2.6.17-ck1.bz2

or server version
http://www.kernel.org/pub/linux/kernel/people/ck/patches/cks/patch-2.6.17-cks1.bz2

web:
http://kernel.kolivas.org

Quoted 8 lines

all patches:
http://www.kernel.org/pub/linux/kernel/people/ck/patches/

Split patches available.

Full patchlist:

<snip>

Thanks Con.
Good news: 2.6.17 has support for the Winbond W83687thf sensor chip. One patch less for me. Bad news: now nvidia 7167 no longer compiles:
*Unknown symbol pm_unregister

Regards,

Hugo

*

#6 lore raw
Quoted 6 lines

<snip>

Thanks Con.
Good news: 2.6.17 has support for the Winbond W83687thf sensor chip. One patch less for me.
Bad news: now nvidia 7167 no longer compiles:

Well that one is old even.

Quoted 1 line

*Unknown symbol pm_unregister

Jan Engelhardt