1 Documentation for /proc/sys/vm/* kernel version 2.6.29
2 (c) 1998, 1999, Rik van Riel <riel@nl.linux.org>
3 (c) 2008 Peter W. Morreale <pmorreale@novell.com>
5 For general info and legal blurb, please look in README.
7 ==============================================================
9 This file contains the documentation for the sysctl files in
10 /proc/sys/vm and is valid for Linux kernel version 2.6.29.
12 The files in this directory can be used to tune the operation
13 of the virtual memory (VM) subsystem of the Linux kernel and
14 the writeout of dirty data to disk.
16 Default values and initialization routines for most of these
17 files can be found in mm/swap.c.
19 Currently, these files are in /proc/sys/vm:
21 - admin_reserve_kbytes
24 - compact_unevictable_allowed
25 - dirty_background_bytes
26 - dirty_background_ratio
28 - dirty_expire_centisecs
30 - dirty_writeback_centisecs
34 - hugepages_treat_as_movable
38 - lowmem_reserve_ratio
40 - memory_failure_early_kill
41 - memory_failure_recovery
47 - nr_overcommit_hugepages
48 - nr_trim_pages (only if CONFIG_MMU=n)
51 - oom_kill_allocating_task
57 - percpu_pagelist_fraction
64 ==============================================================
68 The amount of free memory in the system that should be reserved for users
69 with the capability cap_sys_admin.
71 admin_reserve_kbytes defaults to min(3% of free pages, 8MB)
73 That should provide enough for the admin to log in and kill a process,
74 if necessary, under the default overcommit 'guess' mode.
76 Systems running under overcommit 'never' should increase this to account
77 for the full Virtual Memory Size of programs used to recover. Otherwise,
78 root may not be able to log in to recover the system.
80 How do you calculate a minimum useful reserve?
82 sshd or login + bash (or some other shell) + top (or ps, kill, etc.)
84 For overcommit 'guess', we can sum resident set sizes (RSS).
85 On x86_64 this is about 8MB.
87 For overcommit 'never', we can take the max of their virtual sizes (VSZ)
88 and add the sum of their RSS.
89 On x86_64 this is about 128MB.
91 Changing this takes effect whenever an application requests memory.
93 ==============================================================
97 block_dump enables block I/O debugging when set to a nonzero value. More
98 information on block I/O debugging is in Documentation/laptops/laptop-mode.txt.
100 ==============================================================
104 Available only when CONFIG_COMPACTION is set. When 1 is written to the file,
105 all zones are compacted such that free memory is available in contiguous
106 blocks where possible. This can be important for example in the allocation of
107 huge pages although processes will also directly compact memory as required.
109 ==============================================================
111 compact_unevictable_allowed
113 Available only when CONFIG_COMPACTION is set. When set to 1, compaction is
114 allowed to examine the unevictable lru (mlocked pages) for pages to compact.
115 This should be used on systems where stalls for minor page faults are an
116 acceptable trade for large contiguous free memory. Set to 0 to prevent
117 compaction from moving pages that are unevictable. Default value is 1.
119 ==============================================================
121 dirty_background_bytes
123 Contains the amount of dirty memory at which the background kernel
124 flusher threads will start writeback.
126 Note: dirty_background_bytes is the counterpart of dirty_background_ratio. Only
127 one of them may be specified at a time. When one sysctl is written it is
128 immediately taken into account to evaluate the dirty memory limits and the
129 other appears as 0 when read.
131 ==============================================================
133 dirty_background_ratio
135 Contains, as a percentage of total available memory that contains free pages
136 and reclaimable pages, the number of pages at which the background kernel
137 flusher threads will start writing out dirty data.
139 The total avaiable memory is not equal to total system memory.
141 ==============================================================
145 Contains the amount of dirty memory at which a process generating disk writes
146 will itself start writeback.
148 Note: dirty_bytes is the counterpart of dirty_ratio. Only one of them may be
149 specified at a time. When one sysctl is written it is immediately taken into
150 account to evaluate the dirty memory limits and the other appears as 0 when
153 Note: the minimum value allowed for dirty_bytes is two pages (in bytes); any
154 value lower than this limit will be ignored and the old configuration will be
157 ==============================================================
159 dirty_expire_centisecs
161 This tunable is used to define when dirty data is old enough to be eligible
162 for writeout by the kernel flusher threads. It is expressed in 100'ths
163 of a second. Data which has been dirty in-memory for longer than this
164 interval will be written out next time a flusher thread wakes up.
166 ==============================================================
170 Contains, as a percentage of total available memory that contains free pages
171 and reclaimable pages, the number of pages at which a process which is
172 generating disk writes will itself start writing out dirty data.
174 The total avaiable memory is not equal to total system memory.
176 ==============================================================
178 dirty_writeback_centisecs
180 The kernel flusher threads will periodically wake up and write `old' data
181 out to disk. This tunable expresses the interval between those wakeups, in
184 Setting this to zero disables periodic writeback altogether.
186 ==============================================================
190 Writing to this will cause the kernel to drop clean caches, as well as
191 reclaimable slab objects like dentries and inodes. Once dropped, their
195 echo 1 > /proc/sys/vm/drop_caches
196 To free reclaimable slab objects (includes dentries and inodes):
197 echo 2 > /proc/sys/vm/drop_caches
198 To free slab objects and pagecache:
199 echo 3 > /proc/sys/vm/drop_caches
201 This is a non-destructive operation and will not free any dirty objects.
202 To increase the number of objects freed by this operation, the user may run
203 `sync' prior to writing to /proc/sys/vm/drop_caches. This will minimize the
204 number of dirty objects on the system and create more candidates to be
207 This file is not a means to control the growth of the various kernel caches
208 (inodes, dentries, pagecache, etc...) These objects are automatically
209 reclaimed by the kernel when memory is needed elsewhere on the system.
211 Use of this file can cause performance problems. Since it discards cached
212 objects, it may cost a significant amount of I/O and CPU to recreate the
213 dropped objects, especially if they were under heavy use. Because of this,
214 use outside of a testing or debugging environment is not recommended.
216 You may see informational messages in your kernel log when this file is
219 cat (1234): drop_caches: 3
221 These are informational only. They do not mean that anything is wrong
222 with your system. To disable them, echo 4 (bit 3) into drop_caches.
224 ==============================================================
228 This parameter affects whether the kernel will compact memory or direct
229 reclaim to satisfy a high-order allocation. The extfrag/extfrag_index file in
230 debugfs shows what the fragmentation index for each order is in each zone in
231 the system. Values tending towards 0 imply allocations would fail due to lack
232 of memory, values towards 1000 imply failures are due to fragmentation and -1
233 implies that the allocation will succeed as long as watermarks are met.
235 The kernel will not compact memory in a zone if the
236 fragmentation index is <= extfrag_threshold. The default value is 500.
238 ==============================================================
242 This parameter tells the VM to keep extra free memory between the threshold
243 where background reclaim (kswapd) kicks in, and the threshold where direct
244 reclaim (by allocating processes) kicks in.
246 This is useful for workloads that require low latency memory allocations
247 and have a bounded burstiness in memory allocations, for example a
248 realtime application that receives and transmits network traffic
249 (causing in-kernel memory allocations) with a maximum total message burst
250 size of 200MB may need 200MB of extra free memory to avoid direct reclaim
253 ==============================================================
255 hugepages_treat_as_movable
257 This parameter controls whether we can allocate hugepages from ZONE_MOVABLE
258 or not. If set to non-zero, hugepages can be allocated from ZONE_MOVABLE.
259 ZONE_MOVABLE is created when kernel boot parameter kernelcore= is specified,
260 so this parameter has no effect if used without kernelcore=.
262 Hugepage migration is now available in some situations which depend on the
263 architecture and/or the hugepage size. If a hugepage supports migration,
264 allocation from ZONE_MOVABLE is always enabled for the hugepage regardless
265 of the value of this parameter.
266 IOW, this parameter affects only non-migratable hugepages.
268 Assuming that hugepages are not migratable in your system, one usecase of
269 this parameter is that users can make hugepage pool more extensible by
270 enabling the allocation from ZONE_MOVABLE. This is because on ZONE_MOVABLE
271 page reclaim/migration/compaction work more and you can get contiguous
272 memory more likely. Note that using ZONE_MOVABLE for non-migratable
273 hugepages can do harm to other features like memory hotremove (because
274 memory hotremove expects that memory blocks on ZONE_MOVABLE are always
275 removable,) so it's a trade-off responsible for the users.
277 ==============================================================
281 hugetlb_shm_group contains group id that is allowed to create SysV
282 shared memory segment using hugetlb page.
284 ==============================================================
288 laptop_mode is a knob that controls "laptop mode". All the things that are
289 controlled by this knob are discussed in Documentation/laptops/laptop-mode.txt.
291 ==============================================================
295 If non-zero, this sysctl disables the new 32-bit mmap layout - the kernel
296 will use the legacy (2.4) layout for all processes.
298 ==============================================================
302 For some specialised workloads on highmem machines it is dangerous for
303 the kernel to allow process memory to be allocated from the "lowmem"
304 zone. This is because that memory could then be pinned via the mlock()
305 system call, or by unavailability of swapspace.
307 And on large highmem machines this lack of reclaimable lowmem memory
310 So the Linux page allocator has a mechanism which prevents allocations
311 which _could_ use highmem from using too much lowmem. This means that
312 a certain amount of lowmem is defended from the possibility of being
313 captured into pinned user memory.
315 (The same argument applies to the old 16 megabyte ISA DMA region. This
316 mechanism will also defend that region from allocations which could use
319 The `lowmem_reserve_ratio' tunable determines how aggressive the kernel is
320 in defending these lower zones.
322 If you have a machine which uses highmem or ISA DMA and your
323 applications are using mlock(), or if you are running with no swap then
324 you probably should change the lowmem_reserve_ratio setting.
326 The lowmem_reserve_ratio is an array. You can see them by reading this file.
328 % cat /proc/sys/vm/lowmem_reserve_ratio
331 Note: # of this elements is one fewer than number of zones. Because the highest
332 zone's value is not necessary for following calculation.
334 But, these values are not used directly. The kernel calculates # of protection
335 pages for each zones from them. These are shown as array of protection pages
336 in /proc/zoneinfo like followings. (This is an example of x86-64 box).
337 Each zone has an array of protection pages like this.
348 protection: (0, 2004, 2004, 2004)
349 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
354 These protections are added to score to judge whether this zone should be used
355 for page allocation or should be reclaimed.
357 In this example, if normal pages (index=2) are required to this DMA zone and
358 watermark[WMARK_HIGH] is used for watermark, the kernel judges this zone should
359 not be used because pages_free(1355) is smaller than watermark + protection[2]
360 (4 + 2004 = 2008). If this protection value is 0, this zone would be used for
361 normal page requirement. If requirement is DMA zone(index=0), protection[0]
364 zone[i]'s protection[j] is calculated by following expression.
367 zone[i]->protection[j]
368 = (total sums of managed_pages from zone[i+1] to zone[j] on the node)
369 / lowmem_reserve_ratio[i];
371 (should not be protected. = 0;
373 (not necessary, but looks 0)
375 The default values of lowmem_reserve_ratio[i] are
376 256 (if zone[i] means DMA or DMA32 zone)
378 As above expression, they are reciprocal number of ratio.
379 256 means 1/256. # of protection pages becomes about "0.39%" of total managed
380 pages of higher zones on the node.
382 If you would like to protect more pages, smaller values are effective.
383 The minimum value is 1 (1/1 -> 100%).
385 ==============================================================
389 This file contains the maximum number of memory map areas a process
390 may have. Memory map areas are used as a side-effect of calling
391 malloc, directly by mmap and mprotect, and also when loading shared
394 While most applications need less than a thousand maps, certain
395 programs, particularly malloc debuggers, may consume lots of them,
396 e.g., up to one or two maps per allocation.
398 The default value is 65536.
400 =============================================================
402 memory_failure_early_kill:
404 Control how to kill processes when uncorrected memory error (typically
405 a 2bit error in a memory module) is detected in the background by hardware
406 that cannot be handled by the kernel. In some cases (like the page
407 still having a valid copy on disk) the kernel will handle the failure
408 transparently without affecting any applications. But if there is
409 no other uptodate copy of the data it will kill to prevent any data
410 corruptions from propagating.
412 1: Kill all processes that have the corrupted and not reloadable page mapped
413 as soon as the corruption is detected. Note this is not supported
414 for a few types of pages, like kernel internally allocated data or
415 the swap cache, but works for the majority of user pages.
417 0: Only unmap the corrupted page from all processes and only kill a process
418 who tries to access it.
420 The kill is done using a catchable SIGBUS with BUS_MCEERR_AO, so processes can
421 handle this if they want to.
423 This is only active on architectures/platforms with advanced machine
424 check handling and depends on the hardware capabilities.
426 Applications can override this setting individually with the PR_MCE_KILL prctl
428 ==============================================================
430 memory_failure_recovery
432 Enable memory failure recovery (when supported by the platform)
436 0: Always panic on a memory failure.
438 ==============================================================
442 This is used to force the Linux VM to keep a minimum number
443 of kilobytes free. The VM uses this number to compute a
444 watermark[WMARK_MIN] value for each lowmem zone in the system.
445 Each lowmem zone gets a number of reserved free pages based
446 proportionally on its size.
448 Some minimal amount of memory is needed to satisfy PF_MEMALLOC
449 allocations; if you set this to lower than 1024KB, your system will
450 become subtly broken, and prone to deadlock under high loads.
452 Setting this too high will OOM your machine instantly.
454 =============================================================
458 This is available only on NUMA kernels.
460 A percentage of the total pages in each zone. On Zone reclaim
461 (fallback from the local zone occurs) slabs will be reclaimed if more
462 than this percentage of pages in a zone are reclaimable slab pages.
463 This insures that the slab growth stays under control even in NUMA
464 systems that rarely perform global reclaim.
466 The default is 5 percent.
468 Note that slab reclaim is triggered in a per zone / node fashion.
469 The process of reclaiming slab memory is currently not node specific
472 =============================================================
476 This is available only on NUMA kernels.
478 This is a percentage of the total pages in each zone. Zone reclaim will
479 only occur if more than this percentage of pages are in a state that
480 zone_reclaim_mode allows to be reclaimed.
482 If zone_reclaim_mode has the value 4 OR'd, then the percentage is compared
483 against all file-backed unmapped pages including swapcache pages and tmpfs
484 files. Otherwise, only unmapped pages backed by normal files but not tmpfs
485 files and similar are considered.
487 The default is 1 percent.
489 ==============================================================
493 This file indicates the amount of address space which a user process will
494 be restricted from mmapping. Since kernel null dereference bugs could
495 accidentally operate based on the information in the first couple of pages
496 of memory userspace processes should not be allowed to write to them. By
497 default this value is set to 0 and no protections will be enforced by the
498 security module. Setting this value to something like 64k will allow the
499 vast majority of applications to work correctly and provide defense in depth
500 against future potential kernel bugs.
502 ==============================================================
506 Change the minimum size of the hugepage pool.
508 See Documentation/vm/hugetlbpage.txt
510 ==============================================================
512 nr_overcommit_hugepages
514 Change the maximum size of the hugepage pool. The maximum is
515 nr_hugepages + nr_overcommit_hugepages.
517 See Documentation/vm/hugetlbpage.txt
519 ==============================================================
523 This is available only on NOMMU kernels.
525 This value adjusts the excess page trimming behaviour of power-of-2 aligned
526 NOMMU mmap allocations.
528 A value of 0 disables trimming of allocations entirely, while a value of 1
529 trims excess pages aggressively. Any value >= 1 acts as the watermark where
530 trimming of allocations is initiated.
532 The default value is 1.
534 See Documentation/nommu-mmap.txt for more information.
536 ==============================================================
540 This sysctl is only for NUMA.
541 'where the memory is allocated from' is controlled by zonelists.
542 (This documentation ignores ZONE_HIGHMEM/ZONE_DMA32 for simple explanation.
543 you may be able to read ZONE_DMA as ZONE_DMA32...)
545 In non-NUMA case, a zonelist for GFP_KERNEL is ordered as following.
546 ZONE_NORMAL -> ZONE_DMA
547 This means that a memory allocation request for GFP_KERNEL will
548 get memory from ZONE_DMA only when ZONE_NORMAL is not available.
550 In NUMA case, you can think of following 2 types of order.
551 Assume 2 node NUMA and below is zonelist of Node(0)'s GFP_KERNEL
553 (A) Node(0) ZONE_NORMAL -> Node(0) ZONE_DMA -> Node(1) ZONE_NORMAL
554 (B) Node(0) ZONE_NORMAL -> Node(1) ZONE_NORMAL -> Node(0) ZONE_DMA.
556 Type(A) offers the best locality for processes on Node(0), but ZONE_DMA
557 will be used before ZONE_NORMAL exhaustion. This increases possibility of
558 out-of-memory(OOM) of ZONE_DMA because ZONE_DMA is tend to be small.
560 Type(B) cannot offer the best locality but is more robust against OOM of
563 Type(A) is called as "Node" order. Type (B) is "Zone" order.
565 "Node order" orders the zonelists by node, then by zone within each node.
566 Specify "[Nn]ode" for node order
568 "Zone Order" orders the zonelists by zone type, then by node within each
569 zone. Specify "[Zz]one" for zone order.
571 Specify "[Dd]efault" to request automatic configuration. Autoconfiguration
572 will select "node" order in following case.
573 (1) if the DMA zone does not exist or
574 (2) if the DMA zone comprises greater than 50% of the available memory or
575 (3) if any node's DMA zone comprises greater than 70% of its local memory and
576 the amount of local memory is big enough.
578 Otherwise, "zone" order will be selected. Default order is recommended unless
579 this is causing problems for your system/application.
581 ==============================================================
585 Enables a system-wide task dump (excluding kernel threads) to be produced
586 when the kernel performs an OOM-killing and includes such information as
587 pid, uid, tgid, vm size, rss, nr_ptes, nr_pmds, swapents, oom_score_adj
588 score, and name. This is helpful to determine why the OOM killer was
589 invoked, to identify the rogue task that caused it, and to determine why
590 the OOM killer chose the task it did to kill.
592 If this is set to zero, this information is suppressed. On very
593 large systems with thousands of tasks it may not be feasible to dump
594 the memory state information for each one. Such systems should not
595 be forced to incur a performance penalty in OOM conditions when the
596 information may not be desired.
598 If this is set to non-zero, this information is shown whenever the
599 OOM killer actually kills a memory-hogging task.
601 The default value is 1 (enabled).
603 ==============================================================
605 oom_kill_allocating_task
607 This enables or disables killing the OOM-triggering task in
608 out-of-memory situations.
610 If this is set to zero, the OOM killer will scan through the entire
611 tasklist and select a task based on heuristics to kill. This normally
612 selects a rogue memory-hogging task that frees up a large amount of
615 If this is set to non-zero, the OOM killer simply kills the task that
616 triggered the out-of-memory condition. This avoids the expensive
619 If panic_on_oom is selected, it takes precedence over whatever value
620 is used in oom_kill_allocating_task.
622 The default value is 0.
624 ==============================================================
628 When overcommit_memory is set to 2, the committed address space is not
629 permitted to exceed swap plus this amount of physical RAM. See below.
631 Note: overcommit_kbytes is the counterpart of overcommit_ratio. Only one
632 of them may be specified at a time. Setting one disables the other (which
633 then appears as 0 when read).
635 ==============================================================
639 This value contains a flag that enables memory overcommitment.
641 When this flag is 0, the kernel attempts to estimate the amount
642 of free memory left when userspace requests more memory.
644 When this flag is 1, the kernel pretends there is always enough
645 memory until it actually runs out.
647 When this flag is 2, the kernel uses a "never overcommit"
648 policy that attempts to prevent any overcommit of memory.
649 Note that user_reserve_kbytes affects this policy.
651 This feature can be very useful because there are a lot of
652 programs that malloc() huge amounts of memory "just-in-case"
653 and don't use much of it.
655 The default value is 0.
657 See Documentation/vm/overcommit-accounting and
658 mm/mmap.c::__vm_enough_memory() for more information.
660 ==============================================================
664 When overcommit_memory is set to 2, the committed address
665 space is not permitted to exceed swap plus this percentage
666 of physical RAM. See above.
668 ==============================================================
672 page-cluster controls the number of pages up to which consecutive pages
673 are read in from swap in a single attempt. This is the swap counterpart
674 to page cache readahead.
675 The mentioned consecutivity is not in terms of virtual/physical addresses,
676 but consecutive on swap space - that means they were swapped out together.
678 It is a logarithmic value - setting it to zero means "1 page", setting
679 it to 1 means "2 pages", setting it to 2 means "4 pages", etc.
680 Zero disables swap readahead completely.
682 The default value is three (eight pages at a time). There may be some
683 small benefits in tuning this to a different value if your workload is
686 Lower values mean lower latencies for initial faults, but at the same time
687 extra faults and I/O delays for following faults if they would have been part of
688 that consecutive pages readahead would have brought in.
690 =============================================================
694 This enables or disables panic on out-of-memory feature.
696 If this is set to 0, the kernel will kill some rogue process,
697 called oom_killer. Usually, oom_killer can kill rogue processes and
700 If this is set to 1, the kernel panics when out-of-memory happens.
701 However, if a process limits using nodes by mempolicy/cpusets,
702 and those nodes become memory exhaustion status, one process
703 may be killed by oom-killer. No panic occurs in this case.
704 Because other nodes' memory may be free. This means system total status
705 may be not fatal yet.
707 If this is set to 2, the kernel panics compulsorily even on the
708 above-mentioned. Even oom happens under memory cgroup, the whole
711 The default value is 0.
712 1 and 2 are for failover of clustering. Please select either
713 according to your policy of failover.
714 panic_on_oom=2+kdump gives you very strong tool to investigate
715 why oom happens. You can get snapshot.
717 =============================================================
719 percpu_pagelist_fraction
721 This is the fraction of pages at most (high mark pcp->high) in each zone that
722 are allocated for each per cpu page list. The min value for this is 8. It
723 means that we don't allow more than 1/8th of pages in each zone to be
724 allocated in any single per_cpu_pagelist. This entry only changes the value
725 of hot per cpu pagelists. User can specify a number like 100 to allocate
726 1/100th of each zone to each per cpu page list.
728 The batch value of each per cpu pagelist is also updated as a result. It is
729 set to pcp->high/4. The upper limit of batch is (PAGE_SHIFT * 8)
731 The initial value is zero. Kernel does not use this value at boot time to set
732 the high water marks for each per cpu page list. If the user writes '0' to this
733 sysctl, it will revert to this default behavior.
735 ==============================================================
739 The time interval between which vm statistics are updated. The default
742 ==============================================================
746 This control is used to define how aggressive the kernel will swap
747 memory pages. Higher values will increase agressiveness, lower values
748 decrease the amount of swap. A value of 0 instructs the kernel not to
749 initiate swap until the amount of free and file-backed pages is less
750 than the high water mark in a zone.
752 The default value is 60.
754 ==============================================================
756 - user_reserve_kbytes
758 When overcommit_memory is set to 2, "never overcommit" mode, reserve
759 min(3% of current process size, user_reserve_kbytes) of free memory.
760 This is intended to prevent a user from starting a single memory hogging
761 process, such that they cannot recover (kill the hog).
763 user_reserve_kbytes defaults to min(3% of the current process size, 128MB).
765 If this is reduced to zero, then the user will be allowed to allocate
766 all free memory with a single process, minus admin_reserve_kbytes.
767 Any subsequent attempts to execute a command will result in
768 "fork: Cannot allocate memory".
770 Changing this takes effect whenever an application requests memory.
772 ==============================================================
777 This percentage value controls the tendency of the kernel to reclaim
778 the memory which is used for caching of directory and inode objects.
780 At the default value of vfs_cache_pressure=100 the kernel will attempt to
781 reclaim dentries and inodes at a "fair" rate with respect to pagecache and
782 swapcache reclaim. Decreasing vfs_cache_pressure causes the kernel to prefer
783 to retain dentry and inode caches. When vfs_cache_pressure=0, the kernel will
784 never reclaim dentries and inodes due to memory pressure and this can easily
785 lead to out-of-memory conditions. Increasing vfs_cache_pressure beyond 100
786 causes the kernel to prefer to reclaim dentries and inodes.
788 Increasing vfs_cache_pressure significantly beyond 100 may have negative
789 performance impact. Reclaim code needs to take various locks to find freeable
790 directory and inode objects. With vfs_cache_pressure=1000, it will look for
791 ten times more freeable objects than there are.
793 ==============================================================
797 Zone_reclaim_mode allows someone to set more or less aggressive approaches to
798 reclaim memory when a zone runs out of memory. If it is set to zero then no
799 zone reclaim occurs. Allocations will be satisfied from other zones / nodes
802 This is value ORed together of
805 2 = Zone reclaim writes dirty pages out
806 4 = Zone reclaim swaps pages
808 zone_reclaim_mode is disabled by default. For file servers or workloads
809 that benefit from having their data cached, zone_reclaim_mode should be
810 left disabled as the caching effect is likely to be more important than
813 zone_reclaim may be enabled if it's known that the workload is partitioned
814 such that each partition fits within a NUMA node and that accessing remote
815 memory would cause a measurable performance reduction. The page allocator
816 will then reclaim easily reusable pages (those page cache pages that are
817 currently not used) before allocating off node pages.
819 Allowing zone reclaim to write out pages stops processes that are
820 writing large amounts of data from dirtying pages on other nodes. Zone
821 reclaim will write out dirty pages if a zone fills up and so effectively
822 throttle the process. This may decrease the performance of a single process
823 since it cannot use all of system memory to buffer the outgoing writes
824 anymore but it preserve the memory on other nodes so that the performance
825 of other processes running on other nodes will not be affected.
827 Allowing regular swap effectively restricts allocations to the local
828 node unless explicitly overridden by memory policies or cpuset
831 ============ End of Document =================================