linux/include
Ingo Molnar f6cf891c4d sched: make the scheduler converge to the ideal latency
de-HZ-ification of the granularity defaults unearthed a pre-existing
property of CFS: while it correctly converges to the granularity goal,
it does not prevent run-time fluctuations in the range of
[-gran ... 0 ... +gran].

With the increase of the granularity due to the removal of HZ
dependencies, this becomes visible in chew-max output (with 5 tasks
running):

 out:  28 . 27. 32 | flu:  0 .  0 | ran:    9 .   13 | per:   37 .   40
 out:  27 . 27. 32 | flu:  0 .  0 | ran:   17 .   13 | per:   44 .   40
 out:  27 . 27. 32 | flu:  0 .  0 | ran:    9 .   13 | per:   36 .   40
 out:  29 . 27. 32 | flu:  2 .  0 | ran:   17 .   13 | per:   46 .   40
 out:  28 . 27. 32 | flu:  0 .  0 | ran:    9 .   13 | per:   37 .   40
 out:  29 . 27. 32 | flu:  0 .  0 | ran:   18 .   13 | per:   47 .   40
 out:  28 . 27. 32 | flu:  0 .  0 | ran:    9 .   13 | per:   37 .   40

average slice is the ideal 13 msecs and the period is picture-perfect 40
msecs. But the 'ran' field fluctuates around 13.33 msecs and there's no
mechanism in CFS to keep that from happening: it's a perfectly valid
solution that CFS finds.

to fix this we add a granularity/preemption rule that knows about
the "target latency", which makes tasks that run longer than the ideal
latency run a bit less. The simplest approach is to simply decrease the
preemption granularity when a task overruns its ideal latency. For this
we have to track how much the task executed since its last preemption.

( this adds a new field to task_struct, but we can eliminate that
  overhead in 2.6.24 by putting all the scheduler timestamps into an
  anonymous union. )

with this change in place, chew-max output is fluctuation-less all
around:

 out:  28 . 27. 39 | flu:  0 .  2 | ran:   13 .   13 | per:   41 .   40
 out:  28 . 27. 39 | flu:  0 .  2 | ran:   13 .   13 | per:   41 .   40
 out:  28 . 27. 39 | flu:  0 .  2 | ran:   13 .   13 | per:   41 .   40
 out:  28 . 27. 39 | flu:  0 .  2 | ran:   13 .   13 | per:   41 .   40
 out:  28 . 27. 39 | flu:  0 .  1 | ran:   13 .   13 | per:   41 .   40
 out:  28 . 27. 39 | flu:  0 .  1 | ran:   13 .   13 | per:   41 .   40

this patch has no impact on any fastpath or on any globally observable
scheduling property. (unless you have sharp enough eyes to see
millisecond-level ruckles in glxgears smoothness :-)

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Mike Galbraith <efault@gmx.de>
2007-08-28 12:53:24 +02:00
..
acpi Pull events into release branch 2007-08-25 01:44:01 -04:00
asm-alpha Fix Alpha O_CLOEXEC definition 2007-08-09 08:39:22 -07:00
asm-arm [ARM] 4531/1: remove is_in_rom() protptype 2007-08-04 22:13:43 +01:00
asm-avr32 [AVR32] Define mmiowb() 2007-08-15 16:36:56 +02:00
asm-blackfin remove unused TIF_NOTIFY_RESUME flag 2007-07-31 15:39:38 -07:00
asm-cris remove unused TIF_NOTIFY_RESUME flag 2007-07-31 15:39:38 -07:00
asm-frv FRV: connect up fallocate 2007-08-11 15:47:40 -07:00
asm-generic changing include/asm-generic/pgtable.h for non-mmu 2007-08-11 15:47:42 -07:00
asm-h8300 remove unused TIF_NOTIFY_RESUME flag 2007-07-31 15:39:38 -07:00
asm-i386 Pull bugzilla-1641 into release branch 2007-08-24 22:19:05 -04:00
asm-ia64 [IA64] SN2 needs platform specific irq_to_vector() function. 2007-08-13 10:31:26 -07:00
asm-m32r remove unused TIF_NOTIFY_RESUME flag 2007-07-31 15:39:38 -07:00
asm-m68k m68k: Enable arbitary speed tty support 2007-08-22 19:52:45 -07:00
asm-m68knommu m68knommu: include asm-generic/pgtable.h 2007-08-23 21:32:54 -07:00
asm-mips [MIPS] Polish <asm/edac.h>. 2007-08-27 02:16:59 +01:00
asm-parisc [PARISC] Add dummy isa_(bus|virt)_to_(virt|bus) inlines 2007-08-27 00:29:22 -04:00
asm-powerpc [POWERPC] Fix FSL BookE machine check reporting 2007-08-17 13:22:28 -05:00
asm-ppc [WATCHDOG] mv64x60_wdt: Add arch/powerpc platform support 2007-07-24 21:16:02 +00:00
asm-s390 [S390] Change atomic_read/set to inline functions with barrier semantics. 2007-08-22 13:51:49 +02:00
asm-sh sh: Add missing dma_sync_single_range_for_*(). 2007-08-10 02:37:01 +09:00
asm-sh64 sh64: Add missing dma_sync_single_for_*(). 2007-08-10 02:47:31 +09:00
asm-sparc [SPARC32]: Make flush_tlb_kernel_range() an inline function. 2007-08-26 18:49:12 -07:00
asm-sparc64 [SPARC64]: Need to clobber global reg vars in switch_to(). 2007-08-20 23:40:00 -07:00
asm-um arch: personality independent stack top 2007-07-19 10:04:45 -07:00
asm-v850 remove unused TIF_NOTIFY_RESUME flag 2007-07-31 15:39:38 -07:00
asm-x86_64 Pull bugzilla-1641 into release branch 2007-08-24 22:19:05 -04:00
asm-xtensa remove unused TIF_NOTIFY_RESUME flag 2007-07-31 15:39:38 -07:00
crypto
keys
linux sched: make the scheduler converge to the ideal latency 2007-08-28 12:53:24 +02:00
math-emu Fix <math-emu/soft-fp.h> tpyo 2007-08-18 17:15:17 -07:00
media V4L/DVB (5835): saa7146/dvb-ttpci: Fix signedness warnings (gcc 4.1.1, kernel 2.6.22) 2007-07-18 14:24:44 -03:00
mtd UBI: kill homegrown endian macros 2007-07-18 16:53:49 +03:00
net [NETFILTER]: nf_nat: add symbolic dependency on IPv4 conntrack 2007-08-07 18:12:01 -07:00
pcmcia
rdma IB: Move the macro IB_UMEM_MAX_PAGE_CHUNK() to umem.c 2007-08-03 10:45:18 -07:00
rxrpc
scsi [SCSI] sd: disentangle barriers in SCSI 2007-08-04 08:37:04 -05:00
sound [ALSA] version 1.0.14 2007-07-20 11:13:35 +02:00
video remove tx3912fb 2007-07-31 15:39:41 -07:00
xen xen: xen/page.h compile fix 2007-07-26 11:35:16 -07:00
Kbuild