linux/mm
Christoph Lameter df9ecaba3f [PATCH] ZVC: Scale thresholds depending on the size of the system
The ZVC counter update threshold is currently set to a fixed value of 32.
This patch sets up the threshold depending on the number of processors and
the sizes of the zones in the system.

With the current threshold of 32, I was able to observe slight contention
when more than 130-140 processors concurrently updated the counters.  The
contention vanished when I either increased the threshold to 64 or used
Andrew's idea of overstepping the interval (see ZVC overstep patch).

However, we saw contention again at 220-230 processors.  So we need higher
values for larger systems.

But the current default is already a bit of an overkill for smaller
systems.  Some systems have tiny zones where precision matters.  For
example i386 and x86_64 have 16M DMA zones and either 900M ZONE_NORMAL or
ZONE_DMA32.  These are even present on SMP and NUMA systems.

The patch here sets up a threshold based on the number of processors in the
system and the size of the zone that these counters are used for.  The
threshold should grow logarithmically, so we use fls() as an easy
approximation.

Results of tests on a system with 1024 processors (4TB RAM)

The following output is from a test allocating 1GB of memory concurrently
on each processor (Forking the process.  So contention on mmap_sem and the
pte locks is not a factor):

                       X                   MIN
TYPE:               CPUS       WALL       WALL        SYS     USER     TOTCPU
fork                   1      0.552      0.552      0.540    0.012      0.552
fork                   4      0.552      0.548      2.164    0.036      2.200
fork                  16      0.564      0.548      8.812    0.164      8.976
fork                 128      0.580      0.572     72.204    1.208     73.412
fork                 256      1.300      0.660    310.400    2.160    312.560
fork                 512      3.512      0.696   1526.836    4.816   1531.652
fork                1020     20.024      0.700  17243.176    6.688  17249.863

So a threshold of 32 is fine up to 128 processors. At 256 processors contention
becomes a factor.

Overstepping the counter (earlier patch) improves the numbers a bit:

fork                   4      0.552      0.548      2.164    0.040      2.204
fork                  16      0.552      0.548      8.640    0.148      8.788
fork                 128      0.556      0.548     69.676    0.956     70.632
fork                 256      0.876      0.636    212.468    2.108    214.576
fork                 512      2.276      0.672    997.324    4.260   1001.584
fork                1020     13.564      0.680  11586.436    6.088  11592.523

Still contention at 512 and 1020. Contention at 1020 is down by a third.
256 still has a slight bit of contention.

After this patch the counter threshold will be set to 125 which reduces
contention significantly:

fork                 128      0.560      0.548     69.776    0.932     70.708
fork                 256      0.636      0.556    143.460    2.036    145.496
fork                 512      0.640      0.548    284.244    4.236    288.480
fork                1020      1.500      0.588   1326.152    8.892   1335.044

[akpm@osdl.org: !SMP build fix]
Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-09-01 11:39:08 -07:00
..
bootmem.c [PATCH] mm/bootmem.c: EXPORT_UNUSED_SYMBOL 2006-07-10 13:24:17 -07:00
fadvise.c [PATCH] fadvise() make POSIX_FADV_NOREUSE a no-op 2006-08-06 08:57:47 -07:00
filemap_xip.c [PATCH] mark address_space_operations const 2006-06-28 14:59:04 -07:00
filemap.c [PATCH] MM: Remove rogue readahead printk 2006-07-29 20:59:55 -07:00
filemap.h [PATCH] generic_file_buffered_write(): handle zero-length iovec segments 2006-06-29 10:26:20 -07:00
fremap.c [PATCH] fix update_mmu_cache in fremap.c 2006-06-23 07:42:52 -07:00
highmem.c [PATCH] zoned vm counters: conversion of nr_bounce to per zone counter 2006-06-30 11:25:36 -07:00
hugetlb.c [PATCH] tightening hugetlb strict accounting 2006-06-23 07:42:48 -07:00
internal.h [PATCH] remove set_page_count() outside mm/ 2006-03-22 07:54:02 -08:00
Kconfig Merge master.kernel.org:/pub/scm/linux/kernel/git/gregkh/pci-2.6 2006-06-29 10:49:17 -07:00
madvise.c [PATCH] Fix MADV_REMOVE protection checking 2006-04-17 18:22:18 -07:00
Makefile [PATCH] zoned vm counters: create vmstat.c/.h from page_alloc.c/.h 2006-06-30 11:25:34 -07:00
memory_hotplug.c [PATCH] memory hotadd fixes: enhance collision check 2006-08-06 08:57:49 -07:00
memory.c [PATCH] per-task-delay-accounting: sync block I/O and swapin delay collection 2006-07-14 21:53:56 -07:00
mempolicy.c [PATCH] Use Zoned VM Counters for NUMA statistics 2006-06-30 11:25:36 -07:00
mempool.c Merge git://git.kernel.org/pub/scm/linux/kernel/git/bunk/trivial 2006-03-26 09:41:18 -08:00
migrate.c [PATCH] Allow migration of mlocked pages 2006-06-25 10:00:55 -07:00
mincore.c [PATCH] freepgt: sys_mincore ignore FIRST_USER_PGD_NR 2005-04-19 13:29:20 -07:00
mlock.c [PATCH] move capable() to capability.h 2006-01-11 18:42:13 -08:00
mmap.c [PATCH] zoned vm counters: conversion of nr_pagecache to per zone counter 2006-06-30 11:25:34 -07:00
mmzone.c [PATCH] mm/mmzone.c: EXPORT_UNUSED_SYMBOL 2006-07-10 13:24:17 -07:00
mprotect.c [PATCH] add page_mkwrite() vm_operations method 2006-06-23 07:42:51 -07:00
mremap.c [PATCH] lockdep: annotate mm 2006-07-03 15:27:07 -07:00
msync.c [PATCH] Kill PF_SYNCWRITE flag 2006-06-23 17:10:39 +02:00
nommu.c [PATCH] nommu: export two symbols for drivers to use 2006-07-14 21:53:53 -07:00
oom_kill.c [PATCH] sched: cleanup, remove task_t, convert to struct task_struct 2006-07-03 15:27:11 -07:00
page_alloc.c [PATCH] ZVC/zone_reclaim: Leave 1% of unmapped pagecache pages for file I/O 2006-07-03 15:26:59 -07:00
page_io.c [PATCH] Light weight event counters 2006-06-30 11:25:36 -07:00
page-writeback.c [PATCH] zoned vm counters: remove useless struct wbs 2006-06-30 11:25:36 -07:00
pdflush.c [PATCH] pdflush: handle resume wakeups 2006-06-25 10:01:06 -07:00
prio_tree.c Linux-2.6.12-rc2 2005-04-16 15:20:36 -07:00
readahead.c spelling fixes 2006-06-26 18:35:02 +02:00
rmap.c [PATCH] zoned vm counters: split NR_ANON_PAGES off from NR_FILE_MAPPED 2006-06-30 11:25:35 -07:00
shmem.c Merge git://git.kernel.org/pub/scm/linux/kernel/git/bunk/trivial 2006-06-30 15:39:30 -07:00
slab.c [PATCH] Fix kmem_cache_alloc() been documented twice 2006-07-31 13:28:43 -07:00
slob.c Remove obsolete #include <linux/config.h> 2006-06-30 19:25:36 +02:00
sparse.c Remove obsolete #include <linux/config.h> 2006-06-30 19:25:36 +02:00
swap_state.c [PATCH] lockdep: locking init debugging improvement 2006-07-03 15:27:02 -07:00
swap.c [PATCH] fuse: fix error case in fuse_readpages 2006-08-14 12:54:29 -07:00
swapfile.c [PATCH] swsusp: Fix swap_type_of 2006-08-27 11:01:28 -07:00
thrash.c [PATCH] temporarily disable swap token on memory pressure 2005-11-28 14:42:25 -08:00
tiny-shmem.c [PATCH] devfs: Remove the devfs_fs_kernel.h file from the tree 2006-06-26 12:25:08 -07:00
truncate.c [PATCH] Remove semi-softlockup from invalidate_mapping_pages 2006-06-23 07:43:07 -07:00
util.c [PATCH] slab: optimize constant-size kzalloc calls 2006-03-25 08:22:49 -08:00
vmalloc.c [PATCH] mm: fix oom roll-back of __vmalloc_area_node 2006-07-14 21:53:51 -07:00
vmscan.c [PATCH] ZVC/zone_reclaim: Leave 1% of unmapped pagecache pages for file I/O 2006-07-03 15:26:59 -07:00
vmstat.c [PATCH] ZVC: Scale thresholds depending on the size of the system 2006-09-01 11:39:08 -07:00