1 ===============================
2 Documentation for /proc/sys/vm/
3 ===============================
7 Copyright (c) 1998, 1999, Rik van Riel <riel@nl.linux.org>
9 Copyright (c) 2008 Peter W. Morreale <pmorreale@novell.com>
11 For general info and legal blurb, please look in index.rst.
13 ------------------------------------------------------------------------------
15 This file contains the documentation for the sysctl files in
16 /proc/sys/vm and is valid for Linux kernel version 2.6.29.
18 The files in this directory can be used to tune the operation
19 of the virtual memory (VM) subsystem of the Linux kernel and
20 the writeout of dirty data to disk.
22 Default values and initialization routines for most of these
23 files can be found in mm/swap.c.
25 Currently, these files are in /proc/sys/vm:
27 - admin_reserve_kbytes
30 - compact_unevictable_allowed
31 - dirty_background_bytes
32 - dirty_background_ratio
34 - dirty_expire_centisecs
36 - dirtytime_expire_seconds
37 - dirty_writeback_centisecs
43 - lowmem_reserve_ratio
45 - memory_failure_early_kill
46 - memory_failure_recovery
52 - mmap_rnd_compat_bits
54 - nr_hugepages_mempolicy
55 - nr_overcommit_hugepages
56 - nr_trim_pages (only if CONFIG_MMU=n)
59 - oom_kill_allocating_task
65 - percpu_pagelist_fraction
70 - unprivileged_userfaultfd
73 - watermark_boost_factor
74 - watermark_scale_factor
81 The amount of free memory in the system that should be reserved for users
82 with the capability cap_sys_admin.
84 admin_reserve_kbytes defaults to min(3% of free pages, 8MB)
86 That should provide enough for the admin to log in and kill a process,
87 if necessary, under the default overcommit 'guess' mode.
89 Systems running under overcommit 'never' should increase this to account
90 for the full Virtual Memory Size of programs used to recover. Otherwise,
91 root may not be able to log in to recover the system.
93 How do you calculate a minimum useful reserve?
95 sshd or login + bash (or some other shell) + top (or ps, kill, etc.)
97 For overcommit 'guess', we can sum resident set sizes (RSS).
98 On x86_64 this is about 8MB.
100 For overcommit 'never', we can take the max of their virtual sizes (VSZ)
101 and add the sum of their RSS.
102 On x86_64 this is about 128MB.
104 Changing this takes effect whenever an application requests memory.
110 block_dump enables block I/O debugging when set to a nonzero value. More
111 information on block I/O debugging is in Documentation/admin-guide/laptops/laptop-mode.rst.
117 Available only when CONFIG_COMPACTION is set. When 1 is written to the file,
118 all zones are compacted such that free memory is available in contiguous
119 blocks where possible. This can be important for example in the allocation of
120 huge pages although processes will also directly compact memory as required.
123 compact_unevictable_allowed
124 ===========================
126 Available only when CONFIG_COMPACTION is set. When set to 1, compaction is
127 allowed to examine the unevictable lru (mlocked pages) for pages to compact.
128 This should be used on systems where stalls for minor page faults are an
129 acceptable trade for large contiguous free memory. Set to 0 to prevent
130 compaction from moving pages that are unevictable. Default value is 1.
131 On CONFIG_PREEMPT_RT the default value is 0 in order to avoid a page fault, due
132 to compaction, which would block the task from becomming active until the fault
136 dirty_background_bytes
137 ======================
139 Contains the amount of dirty memory at which the background kernel
140 flusher threads will start writeback.
143 dirty_background_bytes is the counterpart of dirty_background_ratio. Only
144 one of them may be specified at a time. When one sysctl is written it is
145 immediately taken into account to evaluate the dirty memory limits and the
146 other appears as 0 when read.
149 dirty_background_ratio
150 ======================
152 Contains, as a percentage of total available memory that contains free pages
153 and reclaimable pages, the number of pages at which the background kernel
154 flusher threads will start writing out dirty data.
156 The total available memory is not equal to total system memory.
162 Contains the amount of dirty memory at which a process generating disk writes
163 will itself start writeback.
165 Note: dirty_bytes is the counterpart of dirty_ratio. Only one of them may be
166 specified at a time. When one sysctl is written it is immediately taken into
167 account to evaluate the dirty memory limits and the other appears as 0 when
170 Note: the minimum value allowed for dirty_bytes is two pages (in bytes); any
171 value lower than this limit will be ignored and the old configuration will be
175 dirty_expire_centisecs
176 ======================
178 This tunable is used to define when dirty data is old enough to be eligible
179 for writeout by the kernel flusher threads. It is expressed in 100'ths
180 of a second. Data which has been dirty in-memory for longer than this
181 interval will be written out next time a flusher thread wakes up.
187 Contains, as a percentage of total available memory that contains free pages
188 and reclaimable pages, the number of pages at which a process which is
189 generating disk writes will itself start writing out dirty data.
191 The total available memory is not equal to total system memory.
194 dirtytime_expire_seconds
195 ========================
197 When a lazytime inode is constantly having its pages dirtied, the inode with
198 an updated timestamp will never get chance to be written out. And, if the
199 only thing that has happened on the file system is a dirtytime inode caused
200 by an atime update, a worker will be scheduled to make sure that inode
201 eventually gets pushed out to disk. This tunable is used to define when dirty
202 inode is old enough to be eligible for writeback by the kernel flusher threads.
203 And, it is also used as the interval to wakeup dirtytime_writeback thread.
206 dirty_writeback_centisecs
207 =========================
209 The kernel flusher threads will periodically wake up and write `old` data
210 out to disk. This tunable expresses the interval between those wakeups, in
213 Setting this to zero disables periodic writeback altogether.
219 Writing to this will cause the kernel to drop clean caches, as well as
220 reclaimable slab objects like dentries and inodes. Once dropped, their
225 echo 1 > /proc/sys/vm/drop_caches
227 To free reclaimable slab objects (includes dentries and inodes)::
229 echo 2 > /proc/sys/vm/drop_caches
231 To free slab objects and pagecache::
233 echo 3 > /proc/sys/vm/drop_caches
235 This is a non-destructive operation and will not free any dirty objects.
236 To increase the number of objects freed by this operation, the user may run
237 `sync` prior to writing to /proc/sys/vm/drop_caches. This will minimize the
238 number of dirty objects on the system and create more candidates to be
241 This file is not a means to control the growth of the various kernel caches
242 (inodes, dentries, pagecache, etc...) These objects are automatically
243 reclaimed by the kernel when memory is needed elsewhere on the system.
245 Use of this file can cause performance problems. Since it discards cached
246 objects, it may cost a significant amount of I/O and CPU to recreate the
247 dropped objects, especially if they were under heavy use. Because of this,
248 use outside of a testing or debugging environment is not recommended.
250 You may see informational messages in your kernel log when this file is
253 cat (1234): drop_caches: 3
255 These are informational only. They do not mean that anything is wrong
256 with your system. To disable them, echo 4 (bit 2) into drop_caches.
262 This parameter affects whether the kernel will compact memory or direct
263 reclaim to satisfy a high-order allocation. The extfrag/extfrag_index file in
264 debugfs shows what the fragmentation index for each order is in each zone in
265 the system. Values tending towards 0 imply allocations would fail due to lack
266 of memory, values towards 1000 imply failures are due to fragmentation and -1
267 implies that the allocation will succeed as long as watermarks are met.
269 The kernel will not compact memory in a zone if the
270 fragmentation index is <= extfrag_threshold. The default value is 500.
276 Available only for systems with CONFIG_HIGHMEM enabled (32b systems).
278 This parameter controls whether the high memory is considered for dirty
279 writers throttling. This is not the case by default which means that
280 only the amount of memory directly visible/usable by the kernel can
281 be dirtied. As a result, on systems with a large amount of memory and
282 lowmem basically depleted writers might be throttled too early and
283 streaming writes can get very slow.
285 Changing the value to non zero would allow more memory to be dirtied
286 and thus allow writers to write more data which can be flushed to the
287 storage more effectively. Note this also comes with a risk of pre-mature
288 OOM killer because some writers (e.g. direct block device writes) can
289 only use the low memory and they can fill it up with dirty data without
296 hugetlb_shm_group contains group id that is allowed to create SysV
297 shared memory segment using hugetlb page.
303 laptop_mode is a knob that controls "laptop mode". All the things that are
304 controlled by this knob are discussed in Documentation/admin-guide/laptops/laptop-mode.rst.
310 If non-zero, this sysctl disables the new 32-bit mmap layout - the kernel
311 will use the legacy (2.4) layout for all processes.
317 For some specialised workloads on highmem machines it is dangerous for
318 the kernel to allow process memory to be allocated from the "lowmem"
319 zone. This is because that memory could then be pinned via the mlock()
320 system call, or by unavailability of swapspace.
322 And on large highmem machines this lack of reclaimable lowmem memory
325 So the Linux page allocator has a mechanism which prevents allocations
326 which *could* use highmem from using too much lowmem. This means that
327 a certain amount of lowmem is defended from the possibility of being
328 captured into pinned user memory.
330 (The same argument applies to the old 16 megabyte ISA DMA region. This
331 mechanism will also defend that region from allocations which could use
334 The `lowmem_reserve_ratio` tunable determines how aggressive the kernel is
335 in defending these lower zones.
337 If you have a machine which uses highmem or ISA DMA and your
338 applications are using mlock(), or if you are running with no swap then
339 you probably should change the lowmem_reserve_ratio setting.
341 The lowmem_reserve_ratio is an array. You can see them by reading this file::
343 % cat /proc/sys/vm/lowmem_reserve_ratio
346 But, these values are not used directly. The kernel calculates # of protection
347 pages for each zones from them. These are shown as array of protection pages
348 in /proc/zoneinfo like followings. (This is an example of x86-64 box).
349 Each zone has an array of protection pages like this::
359 protection: (0, 2004, 2004, 2004)
360 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
365 These protections are added to score to judge whether this zone should be used
366 for page allocation or should be reclaimed.
368 In this example, if normal pages (index=2) are required to this DMA zone and
369 watermark[WMARK_HIGH] is used for watermark, the kernel judges this zone should
370 not be used because pages_free(1355) is smaller than watermark + protection[2]
371 (4 + 2004 = 2008). If this protection value is 0, this zone would be used for
372 normal page requirement. If requirement is DMA zone(index=0), protection[0]
375 zone[i]'s protection[j] is calculated by following expression::
378 zone[i]->protection[j]
379 = (total sums of managed_pages from zone[i+1] to zone[j] on the node)
380 / lowmem_reserve_ratio[i];
382 (should not be protected. = 0;
384 (not necessary, but looks 0)
386 The default values of lowmem_reserve_ratio[i] are
388 === ====================================
389 256 (if zone[i] means DMA or DMA32 zone)
391 === ====================================
393 As above expression, they are reciprocal number of ratio.
394 256 means 1/256. # of protection pages becomes about "0.39%" of total managed
395 pages of higher zones on the node.
397 If you would like to protect more pages, smaller values are effective.
398 The minimum value is 1 (1/1 -> 100%). The value less than 1 completely
399 disables protection of the pages.
405 This file contains the maximum number of memory map areas a process
406 may have. Memory map areas are used as a side-effect of calling
407 malloc, directly by mmap, mprotect, and madvise, and also when loading
410 While most applications need less than a thousand maps, certain
411 programs, particularly malloc debuggers, may consume lots of them,
412 e.g., up to one or two maps per allocation.
414 The default value is 65536.
417 memory_failure_early_kill:
418 ==========================
420 Control how to kill processes when uncorrected memory error (typically
421 a 2bit error in a memory module) is detected in the background by hardware
422 that cannot be handled by the kernel. In some cases (like the page
423 still having a valid copy on disk) the kernel will handle the failure
424 transparently without affecting any applications. But if there is
425 no other uptodate copy of the data it will kill to prevent any data
426 corruptions from propagating.
428 1: Kill all processes that have the corrupted and not reloadable page mapped
429 as soon as the corruption is detected. Note this is not supported
430 for a few types of pages, like kernel internally allocated data or
431 the swap cache, but works for the majority of user pages.
433 0: Only unmap the corrupted page from all processes and only kill a process
434 who tries to access it.
436 The kill is done using a catchable SIGBUS with BUS_MCEERR_AO, so processes can
437 handle this if they want to.
439 This is only active on architectures/platforms with advanced machine
440 check handling and depends on the hardware capabilities.
442 Applications can override this setting individually with the PR_MCE_KILL prctl
445 memory_failure_recovery
446 =======================
448 Enable memory failure recovery (when supported by the platform)
452 0: Always panic on a memory failure.
458 This is used to force the Linux VM to keep a minimum number
459 of kilobytes free. The VM uses this number to compute a
460 watermark[WMARK_MIN] value for each lowmem zone in the system.
461 Each lowmem zone gets a number of reserved free pages based
462 proportionally on its size.
464 Some minimal amount of memory is needed to satisfy PF_MEMALLOC
465 allocations; if you set this to lower than 1024KB, your system will
466 become subtly broken, and prone to deadlock under high loads.
468 Setting this too high will OOM your machine instantly.
474 This is available only on NUMA kernels.
476 A percentage of the total pages in each zone. On Zone reclaim
477 (fallback from the local zone occurs) slabs will be reclaimed if more
478 than this percentage of pages in a zone are reclaimable slab pages.
479 This insures that the slab growth stays under control even in NUMA
480 systems that rarely perform global reclaim.
482 The default is 5 percent.
484 Note that slab reclaim is triggered in a per zone / node fashion.
485 The process of reclaiming slab memory is currently not node specific
492 This is available only on NUMA kernels.
494 This is a percentage of the total pages in each zone. Zone reclaim will
495 only occur if more than this percentage of pages are in a state that
496 zone_reclaim_mode allows to be reclaimed.
498 If zone_reclaim_mode has the value 4 OR'd, then the percentage is compared
499 against all file-backed unmapped pages including swapcache pages and tmpfs
500 files. Otherwise, only unmapped pages backed by normal files but not tmpfs
501 files and similar are considered.
503 The default is 1 percent.
509 This file indicates the amount of address space which a user process will
510 be restricted from mmapping. Since kernel null dereference bugs could
511 accidentally operate based on the information in the first couple of pages
512 of memory userspace processes should not be allowed to write to them. By
513 default this value is set to 0 and no protections will be enforced by the
514 security module. Setting this value to something like 64k will allow the
515 vast majority of applications to work correctly and provide defense in depth
516 against future potential kernel bugs.
522 This value can be used to select the number of bits to use to
523 determine the random offset to the base address of vma regions
524 resulting from mmap allocations on architectures which support
525 tuning address space randomization. This value will be bounded
526 by the architecture's minimum and maximum supported values.
528 This value can be changed after boot using the
529 /proc/sys/vm/mmap_rnd_bits tunable
535 This value can be used to select the number of bits to use to
536 determine the random offset to the base address of vma regions
537 resulting from mmap allocations for applications run in
538 compatibility mode on architectures which support tuning address
539 space randomization. This value will be bounded by the
540 architecture's minimum and maximum supported values.
542 This value can be changed after boot using the
543 /proc/sys/vm/mmap_rnd_compat_bits tunable
549 Change the minimum size of the hugepage pool.
551 See Documentation/admin-guide/mm/hugetlbpage.rst
554 nr_hugepages_mempolicy
555 ======================
557 Change the size of the hugepage pool at run-time on a specific
560 See Documentation/admin-guide/mm/hugetlbpage.rst
563 nr_overcommit_hugepages
564 =======================
566 Change the maximum size of the hugepage pool. The maximum is
567 nr_hugepages + nr_overcommit_hugepages.
569 See Documentation/admin-guide/mm/hugetlbpage.rst
575 This is available only on NOMMU kernels.
577 This value adjusts the excess page trimming behaviour of power-of-2 aligned
578 NOMMU mmap allocations.
580 A value of 0 disables trimming of allocations entirely, while a value of 1
581 trims excess pages aggressively. Any value >= 1 acts as the watermark where
582 trimming of allocations is initiated.
584 The default value is 1.
586 See Documentation/nommu-mmap.txt for more information.
592 This sysctl is only for NUMA and it is deprecated. Anything but
593 Node order will fail!
595 'where the memory is allocated from' is controlled by zonelists.
597 (This documentation ignores ZONE_HIGHMEM/ZONE_DMA32 for simple explanation.
598 you may be able to read ZONE_DMA as ZONE_DMA32...)
600 In non-NUMA case, a zonelist for GFP_KERNEL is ordered as following.
601 ZONE_NORMAL -> ZONE_DMA
602 This means that a memory allocation request for GFP_KERNEL will
603 get memory from ZONE_DMA only when ZONE_NORMAL is not available.
605 In NUMA case, you can think of following 2 types of order.
606 Assume 2 node NUMA and below is zonelist of Node(0)'s GFP_KERNEL::
608 (A) Node(0) ZONE_NORMAL -> Node(0) ZONE_DMA -> Node(1) ZONE_NORMAL
609 (B) Node(0) ZONE_NORMAL -> Node(1) ZONE_NORMAL -> Node(0) ZONE_DMA.
611 Type(A) offers the best locality for processes on Node(0), but ZONE_DMA
612 will be used before ZONE_NORMAL exhaustion. This increases possibility of
613 out-of-memory(OOM) of ZONE_DMA because ZONE_DMA is tend to be small.
615 Type(B) cannot offer the best locality but is more robust against OOM of
618 Type(A) is called as "Node" order. Type (B) is "Zone" order.
620 "Node order" orders the zonelists by node, then by zone within each node.
621 Specify "[Nn]ode" for node order
623 "Zone Order" orders the zonelists by zone type, then by node within each
624 zone. Specify "[Zz]one" for zone order.
626 Specify "[Dd]efault" to request automatic configuration.
628 On 32-bit, the Normal zone needs to be preserved for allocations accessible
629 by the kernel, so "zone" order will be selected.
631 On 64-bit, devices that require DMA32/DMA are relatively rare, so "node"
632 order will be selected.
634 Default order is recommended unless this is causing problems for your
641 Enables a system-wide task dump (excluding kernel threads) to be produced
642 when the kernel performs an OOM-killing and includes such information as
643 pid, uid, tgid, vm size, rss, pgtables_bytes, swapents, oom_score_adj
644 score, and name. This is helpful to determine why the OOM killer was
645 invoked, to identify the rogue task that caused it, and to determine why
646 the OOM killer chose the task it did to kill.
648 If this is set to zero, this information is suppressed. On very
649 large systems with thousands of tasks it may not be feasible to dump
650 the memory state information for each one. Such systems should not
651 be forced to incur a performance penalty in OOM conditions when the
652 information may not be desired.
654 If this is set to non-zero, this information is shown whenever the
655 OOM killer actually kills a memory-hogging task.
657 The default value is 1 (enabled).
660 oom_kill_allocating_task
661 ========================
663 This enables or disables killing the OOM-triggering task in
664 out-of-memory situations.
666 If this is set to zero, the OOM killer will scan through the entire
667 tasklist and select a task based on heuristics to kill. This normally
668 selects a rogue memory-hogging task that frees up a large amount of
671 If this is set to non-zero, the OOM killer simply kills the task that
672 triggered the out-of-memory condition. This avoids the expensive
675 If panic_on_oom is selected, it takes precedence over whatever value
676 is used in oom_kill_allocating_task.
678 The default value is 0.
684 When overcommit_memory is set to 2, the committed address space is not
685 permitted to exceed swap plus this amount of physical RAM. See below.
687 Note: overcommit_kbytes is the counterpart of overcommit_ratio. Only one
688 of them may be specified at a time. Setting one disables the other (which
689 then appears as 0 when read).
695 This value contains a flag that enables memory overcommitment.
697 When this flag is 0, the kernel attempts to estimate the amount
698 of free memory left when userspace requests more memory.
700 When this flag is 1, the kernel pretends there is always enough
701 memory until it actually runs out.
703 When this flag is 2, the kernel uses a "never overcommit"
704 policy that attempts to prevent any overcommit of memory.
705 Note that user_reserve_kbytes affects this policy.
707 This feature can be very useful because there are a lot of
708 programs that malloc() huge amounts of memory "just-in-case"
709 and don't use much of it.
711 The default value is 0.
713 See Documentation/vm/overcommit-accounting.rst and
714 mm/util.c::__vm_enough_memory() for more information.
720 When overcommit_memory is set to 2, the committed address
721 space is not permitted to exceed swap plus this percentage
722 of physical RAM. See above.
728 page-cluster controls the number of pages up to which consecutive pages
729 are read in from swap in a single attempt. This is the swap counterpart
730 to page cache readahead.
731 The mentioned consecutivity is not in terms of virtual/physical addresses,
732 but consecutive on swap space - that means they were swapped out together.
734 It is a logarithmic value - setting it to zero means "1 page", setting
735 it to 1 means "2 pages", setting it to 2 means "4 pages", etc.
736 Zero disables swap readahead completely.
738 The default value is three (eight pages at a time). There may be some
739 small benefits in tuning this to a different value if your workload is
742 Lower values mean lower latencies for initial faults, but at the same time
743 extra faults and I/O delays for following faults if they would have been part of
744 that consecutive pages readahead would have brought in.
750 This enables or disables panic on out-of-memory feature.
752 If this is set to 0, the kernel will kill some rogue process,
753 called oom_killer. Usually, oom_killer can kill rogue processes and
756 If this is set to 1, the kernel panics when out-of-memory happens.
757 However, if a process limits using nodes by mempolicy/cpusets,
758 and those nodes become memory exhaustion status, one process
759 may be killed by oom-killer. No panic occurs in this case.
760 Because other nodes' memory may be free. This means system total status
761 may be not fatal yet.
763 If this is set to 2, the kernel panics compulsorily even on the
764 above-mentioned. Even oom happens under memory cgroup, the whole
767 The default value is 0.
769 1 and 2 are for failover of clustering. Please select either
770 according to your policy of failover.
772 panic_on_oom=2+kdump gives you very strong tool to investigate
773 why oom happens. You can get snapshot.
776 percpu_pagelist_fraction
777 ========================
779 This is the fraction of pages at most (high mark pcp->high) in each zone that
780 are allocated for each per cpu page list. The min value for this is 8. It
781 means that we don't allow more than 1/8th of pages in each zone to be
782 allocated in any single per_cpu_pagelist. This entry only changes the value
783 of hot per cpu pagelists. User can specify a number like 100 to allocate
784 1/100th of each zone to each per cpu page list.
786 The batch value of each per cpu pagelist is also updated as a result. It is
787 set to pcp->high/4. The upper limit of batch is (PAGE_SHIFT * 8)
789 The initial value is zero. Kernel does not use this value at boot time to set
790 the high water marks for each per cpu page list. If the user writes '0' to this
791 sysctl, it will revert to this default behavior.
797 The time interval between which vm statistics are updated. The default
804 Any read or write (by root only) flushes all the per-cpu vm statistics
805 into their global totals, for more accurate reports when testing
806 e.g. cat /proc/sys/vm/stat_refresh /proc/meminfo
808 As a side-effect, it also checks for negative totals (elsewhere reported
809 as 0) and "fails" with EINVAL if any are found, with a warning in dmesg.
810 (At time of writing, a few stats are known sometimes to be found negative,
811 with no ill effects: errors and warnings on these stats are suppressed.)
817 This interface allows runtime configuration of numa statistics.
819 When page allocation performance becomes a bottleneck and you can tolerate
820 some possible tool breakage and decreased numa counter precision, you can
823 echo 0 > /proc/sys/vm/numa_stat
825 When page allocation performance is not a bottleneck and you want all
826 tooling to work, you can do::
828 echo 1 > /proc/sys/vm/numa_stat
834 This control is used to define how aggressive the kernel will swap
835 memory pages. Higher values will increase aggressiveness, lower values
836 decrease the amount of swap. A value of 0 instructs the kernel not to
837 initiate swap until the amount of free and file-backed pages is less
838 than the high water mark in a zone.
840 The default value is 60.
843 unprivileged_userfaultfd
844 ========================
846 This flag controls whether unprivileged users can use the userfaultfd
847 system calls. Set this to 1 to allow unprivileged users to use the
848 userfaultfd system calls, or set this to 0 to restrict userfaultfd to only
849 privileged users (with SYS_CAP_PTRACE capability).
851 The default value is 1.
857 When overcommit_memory is set to 2, "never overcommit" mode, reserve
858 min(3% of current process size, user_reserve_kbytes) of free memory.
859 This is intended to prevent a user from starting a single memory hogging
860 process, such that they cannot recover (kill the hog).
862 user_reserve_kbytes defaults to min(3% of the current process size, 128MB).
864 If this is reduced to zero, then the user will be allowed to allocate
865 all free memory with a single process, minus admin_reserve_kbytes.
866 Any subsequent attempts to execute a command will result in
867 "fork: Cannot allocate memory".
869 Changing this takes effect whenever an application requests memory.
875 This percentage value controls the tendency of the kernel to reclaim
876 the memory which is used for caching of directory and inode objects.
878 At the default value of vfs_cache_pressure=100 the kernel will attempt to
879 reclaim dentries and inodes at a "fair" rate with respect to pagecache and
880 swapcache reclaim. Decreasing vfs_cache_pressure causes the kernel to prefer
881 to retain dentry and inode caches. When vfs_cache_pressure=0, the kernel will
882 never reclaim dentries and inodes due to memory pressure and this can easily
883 lead to out-of-memory conditions. Increasing vfs_cache_pressure beyond 100
884 causes the kernel to prefer to reclaim dentries and inodes.
886 Increasing vfs_cache_pressure significantly beyond 100 may have negative
887 performance impact. Reclaim code needs to take various locks to find freeable
888 directory and inode objects. With vfs_cache_pressure=1000, it will look for
889 ten times more freeable objects than there are.
892 watermark_boost_factor
893 ======================
895 This factor controls the level of reclaim when memory is being fragmented.
896 It defines the percentage of the high watermark of a zone that will be
897 reclaimed if pages of different mobility are being mixed within pageblocks.
898 The intent is that compaction has less work to do in the future and to
899 increase the success rate of future high-order allocations such as SLUB
900 allocations, THP and hugetlbfs pages.
902 To make it sensible with respect to the watermark_scale_factor
903 parameter, the unit is in fractions of 10,000. The default value of
904 15,000 on !DISCONTIGMEM configurations means that up to 150% of the high
905 watermark will be reclaimed in the event of a pageblock being mixed due
906 to fragmentation. The level of reclaim is determined by the number of
907 fragmentation events that occurred in the recent past. If this value is
908 smaller than a pageblock then a pageblocks worth of pages will be reclaimed
909 (e.g. 2MB on 64-bit x86). A boost factor of 0 will disable the feature.
912 watermark_scale_factor
913 ======================
915 This factor controls the aggressiveness of kswapd. It defines the
916 amount of memory left in a node/system before kswapd is woken up and
917 how much memory needs to be free before kswapd goes back to sleep.
919 The unit is in fractions of 10,000. The default value of 10 means the
920 distances between watermarks are 0.1% of the available memory in the
921 node/system. The maximum value is 1000, or 10% of memory.
923 A high rate of threads entering direct reclaim (allocstall) or kswapd
924 going to sleep prematurely (kswapd_low_wmark_hit_quickly) can indicate
925 that the number of free pages kswapd maintains for latency reasons is
926 too small for the allocation bursts occurring in the system. This knob
927 can then be used to tune kswapd aggressiveness accordingly.
933 Zone_reclaim_mode allows someone to set more or less aggressive approaches to
934 reclaim memory when a zone runs out of memory. If it is set to zero then no
935 zone reclaim occurs. Allocations will be satisfied from other zones / nodes
938 This is value OR'ed together of
940 = ===================================
942 2 Zone reclaim writes dirty pages out
943 4 Zone reclaim swaps pages
944 = ===================================
946 zone_reclaim_mode is disabled by default. For file servers or workloads
947 that benefit from having their data cached, zone_reclaim_mode should be
948 left disabled as the caching effect is likely to be more important than
951 zone_reclaim may be enabled if it's known that the workload is partitioned
952 such that each partition fits within a NUMA node and that accessing remote
953 memory would cause a measurable performance reduction. The page allocator
954 will then reclaim easily reusable pages (those page cache pages that are
955 currently not used) before allocating off node pages.
957 Allowing zone reclaim to write out pages stops processes that are
958 writing large amounts of data from dirtying pages on other nodes. Zone
959 reclaim will write out dirty pages if a zone fills up and so effectively
960 throttle the process. This may decrease the performance of a single process
961 since it cannot use all of system memory to buffer the outgoing writes
962 anymore but it preserve the memory on other nodes so that the performance
963 of other processes running on other nodes will not be affected.
965 Allowing regular swap effectively restricts allocations to the local
966 node unless explicitly overridden by memory policies or cpuset