• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1Documentation 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>
4
5For general info and legal blurb, please look in README.
6
7==============================================================
8
9This file contains the documentation for the sysctl files in
10/proc/sys/vm and is valid for Linux kernel version 2.6.29.
11
12The files in this directory can be used to tune the operation
13of the virtual memory (VM) subsystem of the Linux kernel and
14the writeout of dirty data to disk.
15
16Default values and initialization routines for most of these
17files can be found in mm/swap.c.
18
19Currently, these files are in /proc/sys/vm:
20
21- admin_reserve_kbytes
22- block_dump
23- compact_memory
24- dirty_background_bytes
25- dirty_background_ratio
26- dirty_bytes
27- dirty_expire_centisecs
28- dirty_ratio
29- dirty_writeback_centisecs
30- drop_caches
31- extfrag_threshold
32- extra_free_kbytes
33- hugepages_treat_as_movable
34- hugetlb_shm_group
35- laptop_mode
36- legacy_va_layout
37- lowmem_reserve_ratio
38- max_map_count
39- memory_failure_early_kill
40- memory_failure_recovery
41- min_free_kbytes
42- min_slab_ratio
43- min_unmapped_ratio
44- mmap_min_addr
45- mmap_rnd_bits
46- mmap_rnd_compat_bits
47- nr_hugepages
48- nr_overcommit_hugepages
49- nr_trim_pages         (only if CONFIG_MMU=n)
50- numa_zonelist_order
51- oom_dump_tasks
52- oom_kill_allocating_task
53- overcommit_memory
54- overcommit_ratio
55- page-cluster
56- panic_on_oom
57- percpu_pagelist_fraction
58- stat_interval
59- swappiness
60- user_reserve_kbytes
61- vfs_cache_pressure
62- zone_reclaim_mode
63
64==============================================================
65
66admin_reserve_kbytes
67
68The amount of free memory in the system that should be reserved for users
69with the capability cap_sys_admin.
70
71admin_reserve_kbytes defaults to min(3% of free pages, 8MB)
72
73That should provide enough for the admin to log in and kill a process,
74if necessary, under the default overcommit 'guess' mode.
75
76Systems running under overcommit 'never' should increase this to account
77for the full Virtual Memory Size of programs used to recover. Otherwise,
78root may not be able to log in to recover the system.
79
80How do you calculate a minimum useful reserve?
81
82sshd or login + bash (or some other shell) + top (or ps, kill, etc.)
83
84For overcommit 'guess', we can sum resident set sizes (RSS).
85On x86_64 this is about 8MB.
86
87For overcommit 'never', we can take the max of their virtual sizes (VSZ)
88and add the sum of their RSS.
89On x86_64 this is about 128MB.
90
91Changing this takes effect whenever an application requests memory.
92
93==============================================================
94
95block_dump
96
97block_dump enables block I/O debugging when set to a nonzero value. More
98information on block I/O debugging is in Documentation/laptops/laptop-mode.txt.
99
100==============================================================
101
102compact_memory
103
104Available only when CONFIG_COMPACTION is set. When 1 is written to the file,
105all zones are compacted such that free memory is available in contiguous
106blocks where possible. This can be important for example in the allocation of
107huge pages although processes will also directly compact memory as required.
108
109==============================================================
110
111dirty_background_bytes
112
113Contains the amount of dirty memory at which the background kernel
114flusher threads will start writeback.
115
116Note: dirty_background_bytes is the counterpart of dirty_background_ratio. Only
117one of them may be specified at a time. When one sysctl is written it is
118immediately taken into account to evaluate the dirty memory limits and the
119other appears as 0 when read.
120
121==============================================================
122
123dirty_background_ratio
124
125Contains, as a percentage of total system memory, the number of pages at which
126the background kernel flusher threads will start writing out dirty data.
127
128==============================================================
129
130dirty_bytes
131
132Contains the amount of dirty memory at which a process generating disk writes
133will itself start writeback.
134
135Note: dirty_bytes is the counterpart of dirty_ratio. Only one of them may be
136specified at a time. When one sysctl is written it is immediately taken into
137account to evaluate the dirty memory limits and the other appears as 0 when
138read.
139
140Note: the minimum value allowed for dirty_bytes is two pages (in bytes); any
141value lower than this limit will be ignored and the old configuration will be
142retained.
143
144==============================================================
145
146dirty_expire_centisecs
147
148This tunable is used to define when dirty data is old enough to be eligible
149for writeout by the kernel flusher threads.  It is expressed in 100'ths
150of a second.  Data which has been dirty in-memory for longer than this
151interval will be written out next time a flusher thread wakes up.
152
153==============================================================
154
155dirty_ratio
156
157Contains, as a percentage of total system memory, the number of pages at which
158a process which is generating disk writes will itself start writing out dirty
159data.
160
161==============================================================
162
163dirty_writeback_centisecs
164
165The kernel flusher threads will periodically wake up and write `old' data
166out to disk.  This tunable expresses the interval between those wakeups, in
167100'ths of a second.
168
169Setting this to zero disables periodic writeback altogether.
170
171==============================================================
172
173drop_caches
174
175Writing to this will cause the kernel to drop clean caches, dentries and
176inodes from memory, causing that memory to become free.
177
178To free pagecache:
179	echo 1 > /proc/sys/vm/drop_caches
180To free dentries and inodes:
181	echo 2 > /proc/sys/vm/drop_caches
182To free pagecache, dentries and inodes:
183	echo 3 > /proc/sys/vm/drop_caches
184
185As this is a non-destructive operation and dirty objects are not freeable, the
186user should run `sync' first.
187
188==============================================================
189
190extfrag_threshold
191
192This parameter affects whether the kernel will compact memory or direct
193reclaim to satisfy a high-order allocation. /proc/extfrag_index shows what
194the fragmentation index for each order is in each zone in the system. Values
195tending towards 0 imply allocations would fail due to lack of memory,
196values towards 1000 imply failures are due to fragmentation and -1 implies
197that the allocation will succeed as long as watermarks are met.
198
199The kernel will not compact memory in a zone if the
200fragmentation index is <= extfrag_threshold. The default value is 500.
201
202==============================================================
203
204extra_free_kbytes
205
206This parameter tells the VM to keep extra free memory between the threshold
207where background reclaim (kswapd) kicks in, and the threshold where direct
208reclaim (by allocating processes) kicks in.
209
210This is useful for workloads that require low latency memory allocations
211and have a bounded burstiness in memory allocations, for example a
212realtime application that receives and transmits network traffic
213(causing in-kernel memory allocations) with a maximum total message burst
214size of 200MB may need 200MB of extra free memory to avoid direct reclaim
215related latencies.
216
217==============================================================
218
219hugepages_treat_as_movable
220
221This parameter is only useful when kernelcore= is specified at boot time to
222create ZONE_MOVABLE for pages that may be reclaimed or migrated. Huge pages
223are not movable so are not normally allocated from ZONE_MOVABLE. A non-zero
224value written to hugepages_treat_as_movable allows huge pages to be allocated
225from ZONE_MOVABLE.
226
227Once enabled, the ZONE_MOVABLE is treated as an area of memory the huge
228pages pool can easily grow or shrink within. Assuming that applications are
229not running that mlock() a lot of memory, it is likely the huge pages pool
230can grow to the size of ZONE_MOVABLE by repeatedly entering the desired value
231into nr_hugepages and triggering page reclaim.
232
233==============================================================
234
235hugetlb_shm_group
236
237hugetlb_shm_group contains group id that is allowed to create SysV
238shared memory segment using hugetlb page.
239
240==============================================================
241
242laptop_mode
243
244laptop_mode is a knob that controls "laptop mode". All the things that are
245controlled by this knob are discussed in Documentation/laptops/laptop-mode.txt.
246
247==============================================================
248
249legacy_va_layout
250
251If non-zero, this sysctl disables the new 32-bit mmap layout - the kernel
252will use the legacy (2.4) layout for all processes.
253
254==============================================================
255
256lowmem_reserve_ratio
257
258For some specialised workloads on highmem machines it is dangerous for
259the kernel to allow process memory to be allocated from the "lowmem"
260zone.  This is because that memory could then be pinned via the mlock()
261system call, or by unavailability of swapspace.
262
263And on large highmem machines this lack of reclaimable lowmem memory
264can be fatal.
265
266So the Linux page allocator has a mechanism which prevents allocations
267which _could_ use highmem from using too much lowmem.  This means that
268a certain amount of lowmem is defended from the possibility of being
269captured into pinned user memory.
270
271(The same argument applies to the old 16 megabyte ISA DMA region.  This
272mechanism will also defend that region from allocations which could use
273highmem or lowmem).
274
275The `lowmem_reserve_ratio' tunable determines how aggressive the kernel is
276in defending these lower zones.
277
278If you have a machine which uses highmem or ISA DMA and your
279applications are using mlock(), or if you are running with no swap then
280you probably should change the lowmem_reserve_ratio setting.
281
282The lowmem_reserve_ratio is an array. You can see them by reading this file.
283-
284% cat /proc/sys/vm/lowmem_reserve_ratio
285256     256     32
286-
287Note: # of this elements is one fewer than number of zones. Because the highest
288      zone's value is not necessary for following calculation.
289
290But, these values are not used directly. The kernel calculates # of protection
291pages for each zones from them. These are shown as array of protection pages
292in /proc/zoneinfo like followings. (This is an example of x86-64 box).
293Each zone has an array of protection pages like this.
294
295-
296Node 0, zone      DMA
297  pages free     1355
298        min      3
299        low      3
300        high     4
301	:
302	:
303    numa_other   0
304        protection: (0, 2004, 2004, 2004)
305	^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
306  pagesets
307    cpu: 0 pcp: 0
308        :
309-
310These protections are added to score to judge whether this zone should be used
311for page allocation or should be reclaimed.
312
313In this example, if normal pages (index=2) are required to this DMA zone and
314watermark[WMARK_HIGH] is used for watermark, the kernel judges this zone should
315not be used because pages_free(1355) is smaller than watermark + protection[2]
316(4 + 2004 = 2008). If this protection value is 0, this zone would be used for
317normal page requirement. If requirement is DMA zone(index=0), protection[0]
318(=0) is used.
319
320zone[i]'s protection[j] is calculated by following expression.
321
322(i < j):
323  zone[i]->protection[j]
324  = (total sums of present_pages from zone[i+1] to zone[j] on the node)
325    / lowmem_reserve_ratio[i];
326(i = j):
327   (should not be protected. = 0;
328(i > j):
329   (not necessary, but looks 0)
330
331The default values of lowmem_reserve_ratio[i] are
332    256 (if zone[i] means DMA or DMA32 zone)
333    32  (others).
334As above expression, they are reciprocal number of ratio.
335256 means 1/256. # of protection pages becomes about "0.39%" of total present
336pages of higher zones on the node.
337
338If you would like to protect more pages, smaller values are effective.
339The minimum value is 1 (1/1 -> 100%).
340
341==============================================================
342
343max_map_count:
344
345This file contains the maximum number of memory map areas a process
346may have. Memory map areas are used as a side-effect of calling
347malloc, directly by mmap and mprotect, and also when loading shared
348libraries.
349
350While most applications need less than a thousand maps, certain
351programs, particularly malloc debuggers, may consume lots of them,
352e.g., up to one or two maps per allocation.
353
354The default value is 65536.
355
356=============================================================
357
358memory_failure_early_kill:
359
360Control how to kill processes when uncorrected memory error (typically
361a 2bit error in a memory module) is detected in the background by hardware
362that cannot be handled by the kernel. In some cases (like the page
363still having a valid copy on disk) the kernel will handle the failure
364transparently without affecting any applications. But if there is
365no other uptodate copy of the data it will kill to prevent any data
366corruptions from propagating.
367
3681: Kill all processes that have the corrupted and not reloadable page mapped
369as soon as the corruption is detected.  Note this is not supported
370for a few types of pages, like kernel internally allocated data or
371the swap cache, but works for the majority of user pages.
372
3730: Only unmap the corrupted page from all processes and only kill a process
374who tries to access it.
375
376The kill is done using a catchable SIGBUS with BUS_MCEERR_AO, so processes can
377handle this if they want to.
378
379This is only active on architectures/platforms with advanced machine
380check handling and depends on the hardware capabilities.
381
382Applications can override this setting individually with the PR_MCE_KILL prctl
383
384==============================================================
385
386memory_failure_recovery
387
388Enable memory failure recovery (when supported by the platform)
389
3901: Attempt recovery.
391
3920: Always panic on a memory failure.
393
394==============================================================
395
396min_free_kbytes:
397
398This is used to force the Linux VM to keep a minimum number
399of kilobytes free.  The VM uses this number to compute a
400watermark[WMARK_MIN] value for each lowmem zone in the system.
401Each lowmem zone gets a number of reserved free pages based
402proportionally on its size.
403
404Some minimal amount of memory is needed to satisfy PF_MEMALLOC
405allocations; if you set this to lower than 1024KB, your system will
406become subtly broken, and prone to deadlock under high loads.
407
408Setting this too high will OOM your machine instantly.
409
410=============================================================
411
412min_slab_ratio:
413
414This is available only on NUMA kernels.
415
416A percentage of the total pages in each zone.  On Zone reclaim
417(fallback from the local zone occurs) slabs will be reclaimed if more
418than this percentage of pages in a zone are reclaimable slab pages.
419This insures that the slab growth stays under control even in NUMA
420systems that rarely perform global reclaim.
421
422The default is 5 percent.
423
424Note that slab reclaim is triggered in a per zone / node fashion.
425The process of reclaiming slab memory is currently not node specific
426and may not be fast.
427
428=============================================================
429
430min_unmapped_ratio:
431
432This is available only on NUMA kernels.
433
434This is a percentage of the total pages in each zone. Zone reclaim will
435only occur if more than this percentage of pages are in a state that
436zone_reclaim_mode allows to be reclaimed.
437
438If zone_reclaim_mode has the value 4 OR'd, then the percentage is compared
439against all file-backed unmapped pages including swapcache pages and tmpfs
440files. Otherwise, only unmapped pages backed by normal files but not tmpfs
441files and similar are considered.
442
443The default is 1 percent.
444
445==============================================================
446
447mmap_min_addr
448
449This file indicates the amount of address space  which a user process will
450be restricted from mmapping.  Since kernel null dereference bugs could
451accidentally operate based on the information in the first couple of pages
452of memory userspace processes should not be allowed to write to them.  By
453default this value is set to 0 and no protections will be enforced by the
454security module.  Setting this value to something like 64k will allow the
455vast majority of applications to work correctly and provide defense in depth
456against future potential kernel bugs.
457
458==============================================================
459
460mmap_rnd_bits:
461
462This value can be used to select the number of bits to use to
463determine the random offset to the base address of vma regions
464resulting from mmap allocations on architectures which support
465tuning address space randomization.  This value will be bounded
466by the architecture's minimum and maximum supported values.
467
468This value can be changed after boot using the
469/proc/sys/vm/mmap_rnd_bits tunable
470
471==============================================================
472
473mmap_rnd_compat_bits:
474
475This value can be used to select the number of bits to use to
476determine the random offset to the base address of vma regions
477resulting from mmap allocations for applications run in
478compatibility mode on architectures which support tuning address
479space randomization.  This value will be bounded by the
480architecture's minimum and maximum supported values.
481
482This value can be changed after boot using the
483/proc/sys/vm/mmap_rnd_compat_bits tunable
484
485==============================================================
486
487nr_hugepages
488
489Change the minimum size of the hugepage pool.
490
491See Documentation/vm/hugetlbpage.txt
492
493==============================================================
494
495nr_overcommit_hugepages
496
497Change the maximum size of the hugepage pool. The maximum is
498nr_hugepages + nr_overcommit_hugepages.
499
500See Documentation/vm/hugetlbpage.txt
501
502==============================================================
503
504nr_trim_pages
505
506This is available only on NOMMU kernels.
507
508This value adjusts the excess page trimming behaviour of power-of-2 aligned
509NOMMU mmap allocations.
510
511A value of 0 disables trimming of allocations entirely, while a value of 1
512trims excess pages aggressively. Any value >= 1 acts as the watermark where
513trimming of allocations is initiated.
514
515The default value is 1.
516
517See Documentation/nommu-mmap.txt for more information.
518
519==============================================================
520
521numa_zonelist_order
522
523This sysctl is only for NUMA.
524'where the memory is allocated from' is controlled by zonelists.
525(This documentation ignores ZONE_HIGHMEM/ZONE_DMA32 for simple explanation.
526 you may be able to read ZONE_DMA as ZONE_DMA32...)
527
528In non-NUMA case, a zonelist for GFP_KERNEL is ordered as following.
529ZONE_NORMAL -> ZONE_DMA
530This means that a memory allocation request for GFP_KERNEL will
531get memory from ZONE_DMA only when ZONE_NORMAL is not available.
532
533In NUMA case, you can think of following 2 types of order.
534Assume 2 node NUMA and below is zonelist of Node(0)'s GFP_KERNEL
535
536(A) Node(0) ZONE_NORMAL -> Node(0) ZONE_DMA -> Node(1) ZONE_NORMAL
537(B) Node(0) ZONE_NORMAL -> Node(1) ZONE_NORMAL -> Node(0) ZONE_DMA.
538
539Type(A) offers the best locality for processes on Node(0), but ZONE_DMA
540will be used before ZONE_NORMAL exhaustion. This increases possibility of
541out-of-memory(OOM) of ZONE_DMA because ZONE_DMA is tend to be small.
542
543Type(B) cannot offer the best locality but is more robust against OOM of
544the DMA zone.
545
546Type(A) is called as "Node" order. Type (B) is "Zone" order.
547
548"Node order" orders the zonelists by node, then by zone within each node.
549Specify "[Nn]ode" for node order
550
551"Zone Order" orders the zonelists by zone type, then by node within each
552zone.  Specify "[Zz]one" for zone order.
553
554Specify "[Dd]efault" to request automatic configuration.  Autoconfiguration
555will select "node" order in following case.
556(1) if the DMA zone does not exist or
557(2) if the DMA zone comprises greater than 50% of the available memory or
558(3) if any node's DMA zone comprises greater than 60% of its local memory and
559    the amount of local memory is big enough.
560
561Otherwise, "zone" order will be selected. Default order is recommended unless
562this is causing problems for your system/application.
563
564==============================================================
565
566oom_dump_tasks
567
568Enables a system-wide task dump (excluding kernel threads) to be
569produced when the kernel performs an OOM-killing and includes such
570information as pid, uid, tgid, vm size, rss, nr_ptes, swapents,
571oom_score_adj score, and name.  This is helpful to determine why the
572OOM killer was invoked, to identify the rogue task that caused it,
573and to determine why the OOM killer chose the task it did to kill.
574
575If this is set to zero, this information is suppressed.  On very
576large systems with thousands of tasks it may not be feasible to dump
577the memory state information for each one.  Such systems should not
578be forced to incur a performance penalty in OOM conditions when the
579information may not be desired.
580
581If this is set to non-zero, this information is shown whenever the
582OOM killer actually kills a memory-hogging task.
583
584The default value is 1 (enabled).
585
586==============================================================
587
588oom_kill_allocating_task
589
590This enables or disables killing the OOM-triggering task in
591out-of-memory situations.
592
593If this is set to zero, the OOM killer will scan through the entire
594tasklist and select a task based on heuristics to kill.  This normally
595selects a rogue memory-hogging task that frees up a large amount of
596memory when killed.
597
598If this is set to non-zero, the OOM killer simply kills the task that
599triggered the out-of-memory condition.  This avoids the expensive
600tasklist scan.
601
602If panic_on_oom is selected, it takes precedence over whatever value
603is used in oom_kill_allocating_task.
604
605The default value is 0.
606
607==============================================================
608
609overcommit_memory:
610
611This value contains a flag that enables memory overcommitment.
612
613When this flag is 0, the kernel attempts to estimate the amount
614of free memory left when userspace requests more memory.
615
616When this flag is 1, the kernel pretends there is always enough
617memory until it actually runs out.
618
619When this flag is 2, the kernel uses a "never overcommit"
620policy that attempts to prevent any overcommit of memory.
621Note that user_reserve_kbytes affects this policy.
622
623This feature can be very useful because there are a lot of
624programs that malloc() huge amounts of memory "just-in-case"
625and don't use much of it.
626
627The default value is 0.
628
629See Documentation/vm/overcommit-accounting and
630security/commoncap.c::cap_vm_enough_memory() for more information.
631
632==============================================================
633
634overcommit_ratio:
635
636When overcommit_memory is set to 2, the committed address
637space is not permitted to exceed swap plus this percentage
638of physical RAM.  See above.
639
640==============================================================
641
642page-cluster
643
644page-cluster controls the number of pages up to which consecutive pages
645are read in from swap in a single attempt. This is the swap counterpart
646to page cache readahead.
647The mentioned consecutivity is not in terms of virtual/physical addresses,
648but consecutive on swap space - that means they were swapped out together.
649
650It is a logarithmic value - setting it to zero means "1 page", setting
651it to 1 means "2 pages", setting it to 2 means "4 pages", etc.
652Zero disables swap readahead completely.
653
654The default value is three (eight pages at a time).  There may be some
655small benefits in tuning this to a different value if your workload is
656swap-intensive.
657
658Lower values mean lower latencies for initial faults, but at the same time
659extra faults and I/O delays for following faults if they would have been part of
660that consecutive pages readahead would have brought in.
661
662=============================================================
663
664panic_on_oom
665
666This enables or disables panic on out-of-memory feature.
667
668If this is set to 0, the kernel will kill some rogue process,
669called oom_killer.  Usually, oom_killer can kill rogue processes and
670system will survive.
671
672If this is set to 1, the kernel panics when out-of-memory happens.
673However, if a process limits using nodes by mempolicy/cpusets,
674and those nodes become memory exhaustion status, one process
675may be killed by oom-killer. No panic occurs in this case.
676Because other nodes' memory may be free. This means system total status
677may be not fatal yet.
678
679If this is set to 2, the kernel panics compulsorily even on the
680above-mentioned. Even oom happens under memory cgroup, the whole
681system panics.
682
683The default value is 0.
6841 and 2 are for failover of clustering. Please select either
685according to your policy of failover.
686panic_on_oom=2+kdump gives you very strong tool to investigate
687why oom happens. You can get snapshot.
688
689=============================================================
690
691percpu_pagelist_fraction
692
693This is the fraction of pages at most (high mark pcp->high) in each zone that
694are allocated for each per cpu page list.  The min value for this is 8.  It
695means that we don't allow more than 1/8th of pages in each zone to be
696allocated in any single per_cpu_pagelist.  This entry only changes the value
697of hot per cpu pagelists.  User can specify a number like 100 to allocate
6981/100th of each zone to each per cpu page list.
699
700The batch value of each per cpu pagelist is also updated as a result.  It is
701set to pcp->high/4.  The upper limit of batch is (PAGE_SHIFT * 8)
702
703The initial value is zero.  Kernel does not use this value at boot time to set
704the high water marks for each per cpu page list.
705
706==============================================================
707
708stat_interval
709
710The time interval between which vm statistics are updated.  The default
711is 1 second.
712
713==============================================================
714
715swappiness
716
717This control is used to define how aggressive the kernel will swap
718memory pages.  Higher values will increase agressiveness, lower values
719decrease the amount of swap.
720
721The default value is 60.
722
723==============================================================
724
725- user_reserve_kbytes
726
727When overcommit_memory is set to 2, "never overommit" mode, reserve
728min(3% of current process size, user_reserve_kbytes) of free memory.
729This is intended to prevent a user from starting a single memory hogging
730process, such that they cannot recover (kill the hog).
731
732user_reserve_kbytes defaults to min(3% of the current process size, 128MB).
733
734If this is reduced to zero, then the user will be allowed to allocate
735all free memory with a single process, minus admin_reserve_kbytes.
736Any subsequent attempts to execute a command will result in
737"fork: Cannot allocate memory".
738
739Changing this takes effect whenever an application requests memory.
740
741==============================================================
742
743vfs_cache_pressure
744------------------
745
746Controls the tendency of the kernel to reclaim the memory which is used for
747caching of directory and inode objects.
748
749At the default value of vfs_cache_pressure=100 the kernel will attempt to
750reclaim dentries and inodes at a "fair" rate with respect to pagecache and
751swapcache reclaim.  Decreasing vfs_cache_pressure causes the kernel to prefer
752to retain dentry and inode caches. When vfs_cache_pressure=0, the kernel will
753never reclaim dentries and inodes due to memory pressure and this can easily
754lead to out-of-memory conditions. Increasing vfs_cache_pressure beyond 100
755causes the kernel to prefer to reclaim dentries and inodes.
756
757==============================================================
758
759zone_reclaim_mode:
760
761Zone_reclaim_mode allows someone to set more or less aggressive approaches to
762reclaim memory when a zone runs out of memory. If it is set to zero then no
763zone reclaim occurs. Allocations will be satisfied from other zones / nodes
764in the system.
765
766This is value ORed together of
767
7681	= Zone reclaim on
7692	= Zone reclaim writes dirty pages out
7704	= Zone reclaim swaps pages
771
772zone_reclaim_mode is set during bootup to 1 if it is determined that pages
773from remote zones will cause a measurable performance reduction. The
774page allocator will then reclaim easily reusable pages (those page
775cache pages that are currently not used) before allocating off node pages.
776
777It may be beneficial to switch off zone reclaim if the system is
778used for a file server and all of memory should be used for caching files
779from disk. In that case the caching effect is more important than
780data locality.
781
782Allowing zone reclaim to write out pages stops processes that are
783writing large amounts of data from dirtying pages on other nodes. Zone
784reclaim will write out dirty pages if a zone fills up and so effectively
785throttle the process. This may decrease the performance of a single process
786since it cannot use all of system memory to buffer the outgoing writes
787anymore but it preserve the memory on other nodes so that the performance
788of other processes running on other nodes will not be affected.
789
790Allowing regular swap effectively restricts allocations to the local
791node unless explicitly overridden by memory policies or cpuset
792configurations.
793
794============ End of Document =================================
795