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
29 - compaction_proactiveness
30 - compact_unevictable_allowed
31 - dirty_background_bytes
32 - dirty_background_ratio
34 - dirty_expire_centisecs
36 - dirtytime_expire_seconds
37 - dirty_writeback_centisecs
40 - highmem_is_dirtyable
44 - lowmem_reserve_ratio
46 - memory_failure_early_kill
47 - memory_failure_recovery
53 - mmap_rnd_compat_bits
55 - nr_hugepages_mempolicy
56 - nr_overcommit_hugepages
57 - nr_trim_pages (only if CONFIG_MMU=n)
60 - oom_kill_allocating_task
65 - page_lock_unfairness
67 - percpu_pagelist_high_fraction
72 - unprivileged_userfaultfd
75 - watermark_boost_factor
76 - watermark_scale_factor
83 The amount of free memory in the system that should be reserved for users
84 with the capability cap_sys_admin.
86 admin_reserve_kbytes defaults to min(3% of free pages, 8MB)
88 That should provide enough for the admin to log in and kill a process,
89 if necessary, under the default overcommit 'guess' mode.
91 Systems running under overcommit 'never' should increase this to account
92 for the full Virtual Memory Size of programs used to recover. Otherwise,
93 root may not be able to log in to recover the system.
95 How do you calculate a minimum useful reserve?
97 sshd or login + bash (or some other shell) + top (or ps, kill, etc.)
99 For overcommit 'guess', we can sum resident set sizes (RSS).
100 On x86_64 this is about 8MB.
102 For overcommit 'never', we can take the max of their virtual sizes (VSZ)
103 and add the sum of their RSS.
104 On x86_64 this is about 128MB.
106 Changing this takes effect whenever an application requests memory.
112 Available only when CONFIG_COMPACTION is set. When 1 is written to the file,
113 all zones are compacted such that free memory is available in contiguous
114 blocks where possible. This can be important for example in the allocation of
115 huge pages although processes will also directly compact memory as required.
117 compaction_proactiveness
118 ========================
120 This tunable takes a value in the range [0, 100] with a default value of
121 20. This tunable determines how aggressively compaction is done in the
122 background. Write of a non zero value to this tunable will immediately
123 trigger the proactive compaction. Setting it to 0 disables proactive compaction.
125 Note that compaction has a non-trivial system-wide impact as pages
126 belonging to different processes are moved around, which could also lead
127 to latency spikes in unsuspecting applications. The kernel employs
128 various heuristics to avoid wasting CPU cycles if it detects that
129 proactive compaction is not being effective.
131 Be careful when setting it to extreme values like 100, as that may
132 cause excessive background compaction activity.
134 compact_unevictable_allowed
135 ===========================
137 Available only when CONFIG_COMPACTION is set. When set to 1, compaction is
138 allowed to examine the unevictable lru (mlocked pages) for pages to compact.
139 This should be used on systems where stalls for minor page faults are an
140 acceptable trade for large contiguous free memory. Set to 0 to prevent
141 compaction from moving pages that are unevictable. Default value is 1.
142 On CONFIG_PREEMPT_RT the default value is 0 in order to avoid a page fault, due
143 to compaction, which would block the task from becoming active until the fault
147 dirty_background_bytes
148 ======================
150 Contains the amount of dirty memory at which the background kernel
151 flusher threads will start writeback.
154 dirty_background_bytes is the counterpart of dirty_background_ratio. Only
155 one of them may be specified at a time. When one sysctl is written it is
156 immediately taken into account to evaluate the dirty memory limits and the
157 other appears as 0 when read.
160 dirty_background_ratio
161 ======================
163 Contains, as a percentage of total available memory that contains free pages
164 and reclaimable pages, the number of pages at which the background kernel
165 flusher threads will start writing out dirty data.
167 The total available memory is not equal to total system memory.
173 Contains the amount of dirty memory at which a process generating disk writes
174 will itself start writeback.
176 Note: dirty_bytes is the counterpart of dirty_ratio. Only one of them may be
177 specified at a time. When one sysctl is written it is immediately taken into
178 account to evaluate the dirty memory limits and the other appears as 0 when
181 Note: the minimum value allowed for dirty_bytes is two pages (in bytes); any
182 value lower than this limit will be ignored and the old configuration will be
186 dirty_expire_centisecs
187 ======================
189 This tunable is used to define when dirty data is old enough to be eligible
190 for writeout by the kernel flusher threads. It is expressed in 100'ths
191 of a second. Data which has been dirty in-memory for longer than this
192 interval will be written out next time a flusher thread wakes up.
198 Contains, as a percentage of total available memory that contains free pages
199 and reclaimable pages, the number of pages at which a process which is
200 generating disk writes will itself start writing out dirty data.
202 The total available memory is not equal to total system memory.
205 dirtytime_expire_seconds
206 ========================
208 When a lazytime inode is constantly having its pages dirtied, the inode with
209 an updated timestamp will never get chance to be written out. And, if the
210 only thing that has happened on the file system is a dirtytime inode caused
211 by an atime update, a worker will be scheduled to make sure that inode
212 eventually gets pushed out to disk. This tunable is used to define when dirty
213 inode is old enough to be eligible for writeback by the kernel flusher threads.
214 And, it is also used as the interval to wakeup dirtytime_writeback thread.
217 dirty_writeback_centisecs
218 =========================
220 The kernel flusher threads will periodically wake up and write `old` data
221 out to disk. This tunable expresses the interval between those wakeups, in
224 Setting this to zero disables periodic writeback altogether.
230 Writing to this will cause the kernel to drop clean caches, as well as
231 reclaimable slab objects like dentries and inodes. Once dropped, their
236 echo 1 > /proc/sys/vm/drop_caches
238 To free reclaimable slab objects (includes dentries and inodes)::
240 echo 2 > /proc/sys/vm/drop_caches
242 To free slab objects and pagecache::
244 echo 3 > /proc/sys/vm/drop_caches
246 This is a non-destructive operation and will not free any dirty objects.
247 To increase the number of objects freed by this operation, the user may run
248 `sync` prior to writing to /proc/sys/vm/drop_caches. This will minimize the
249 number of dirty objects on the system and create more candidates to be
252 This file is not a means to control the growth of the various kernel caches
253 (inodes, dentries, pagecache, etc...) These objects are automatically
254 reclaimed by the kernel when memory is needed elsewhere on the system.
256 Use of this file can cause performance problems. Since it discards cached
257 objects, it may cost a significant amount of I/O and CPU to recreate the
258 dropped objects, especially if they were under heavy use. Because of this,
259 use outside of a testing or debugging environment is not recommended.
261 You may see informational messages in your kernel log when this file is
264 cat (1234): drop_caches: 3
266 These are informational only. They do not mean that anything is wrong
267 with your system. To disable them, echo 4 (bit 2) into drop_caches.
273 This parameter affects whether the kernel will compact memory or direct
274 reclaim to satisfy a high-order allocation. The extfrag/extfrag_index file in
275 debugfs shows what the fragmentation index for each order is in each zone in
276 the system. Values tending towards 0 imply allocations would fail due to lack
277 of memory, values towards 1000 imply failures are due to fragmentation and -1
278 implies that the allocation will succeed as long as watermarks are met.
280 The kernel will not compact memory in a zone if the
281 fragmentation index is <= extfrag_threshold. The default value is 500.
287 Available only for systems with CONFIG_HIGHMEM enabled (32b systems).
289 This parameter controls whether the high memory is considered for dirty
290 writers throttling. This is not the case by default which means that
291 only the amount of memory directly visible/usable by the kernel can
292 be dirtied. As a result, on systems with a large amount of memory and
293 lowmem basically depleted writers might be throttled too early and
294 streaming writes can get very slow.
296 Changing the value to non zero would allow more memory to be dirtied
297 and thus allow writers to write more data which can be flushed to the
298 storage more effectively. Note this also comes with a risk of pre-mature
299 OOM killer because some writers (e.g. direct block device writes) can
300 only use the low memory and they can fill it up with dirty data without
307 hugetlb_shm_group contains group id that is allowed to create SysV
308 shared memory segment using hugetlb page.
314 laptop_mode is a knob that controls "laptop mode". All the things that are
315 controlled by this knob are discussed in Documentation/admin-guide/laptops/laptop-mode.rst.
321 If non-zero, this sysctl disables the new 32-bit mmap layout - the kernel
322 will use the legacy (2.4) layout for all processes.
328 For some specialised workloads on highmem machines it is dangerous for
329 the kernel to allow process memory to be allocated from the "lowmem"
330 zone. This is because that memory could then be pinned via the mlock()
331 system call, or by unavailability of swapspace.
333 And on large highmem machines this lack of reclaimable lowmem memory
336 So the Linux page allocator has a mechanism which prevents allocations
337 which *could* use highmem from using too much lowmem. This means that
338 a certain amount of lowmem is defended from the possibility of being
339 captured into pinned user memory.
341 (The same argument applies to the old 16 megabyte ISA DMA region. This
342 mechanism will also defend that region from allocations which could use
345 The `lowmem_reserve_ratio` tunable determines how aggressive the kernel is
346 in defending these lower zones.
348 If you have a machine which uses highmem or ISA DMA and your
349 applications are using mlock(), or if you are running with no swap then
350 you probably should change the lowmem_reserve_ratio setting.
352 The lowmem_reserve_ratio is an array. You can see them by reading this file::
354 % cat /proc/sys/vm/lowmem_reserve_ratio
357 But, these values are not used directly. The kernel calculates # of protection
358 pages for each zones from them. These are shown as array of protection pages
359 in /proc/zoneinfo like followings. (This is an example of x86-64 box).
360 Each zone has an array of protection pages like this::
370 protection: (0, 2004, 2004, 2004)
371 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
376 These protections are added to score to judge whether this zone should be used
377 for page allocation or should be reclaimed.
379 In this example, if normal pages (index=2) are required to this DMA zone and
380 watermark[WMARK_HIGH] is used for watermark, the kernel judges this zone should
381 not be used because pages_free(1355) is smaller than watermark + protection[2]
382 (4 + 2004 = 2008). If this protection value is 0, this zone would be used for
383 normal page requirement. If requirement is DMA zone(index=0), protection[0]
386 zone[i]'s protection[j] is calculated by following expression::
389 zone[i]->protection[j]
390 = (total sums of managed_pages from zone[i+1] to zone[j] on the node)
391 / lowmem_reserve_ratio[i];
393 (should not be protected. = 0;
395 (not necessary, but looks 0)
397 The default values of lowmem_reserve_ratio[i] are
399 === ====================================
400 256 (if zone[i] means DMA or DMA32 zone)
402 === ====================================
404 As above expression, they are reciprocal number of ratio.
405 256 means 1/256. # of protection pages becomes about "0.39%" of total managed
406 pages of higher zones on the node.
408 If you would like to protect more pages, smaller values are effective.
409 The minimum value is 1 (1/1 -> 100%). The value less than 1 completely
410 disables protection of the pages.
416 This file contains the maximum number of memory map areas a process
417 may have. Memory map areas are used as a side-effect of calling
418 malloc, directly by mmap, mprotect, and madvise, and also when loading
421 While most applications need less than a thousand maps, certain
422 programs, particularly malloc debuggers, may consume lots of them,
423 e.g., up to one or two maps per allocation.
425 The default value is 65530.
428 memory_failure_early_kill:
429 ==========================
431 Control how to kill processes when uncorrected memory error (typically
432 a 2bit error in a memory module) is detected in the background by hardware
433 that cannot be handled by the kernel. In some cases (like the page
434 still having a valid copy on disk) the kernel will handle the failure
435 transparently without affecting any applications. But if there is
436 no other uptodate copy of the data it will kill to prevent any data
437 corruptions from propagating.
439 1: Kill all processes that have the corrupted and not reloadable page mapped
440 as soon as the corruption is detected. Note this is not supported
441 for a few types of pages, like kernel internally allocated data or
442 the swap cache, but works for the majority of user pages.
444 0: Only unmap the corrupted page from all processes and only kill a process
445 who tries to access it.
447 The kill is done using a catchable SIGBUS with BUS_MCEERR_AO, so processes can
448 handle this if they want to.
450 This is only active on architectures/platforms with advanced machine
451 check handling and depends on the hardware capabilities.
453 Applications can override this setting individually with the PR_MCE_KILL prctl
456 memory_failure_recovery
457 =======================
459 Enable memory failure recovery (when supported by the platform)
463 0: Always panic on a memory failure.
469 This is used to force the Linux VM to keep a minimum number
470 of kilobytes free. The VM uses this number to compute a
471 watermark[WMARK_MIN] value for each lowmem zone in the system.
472 Each lowmem zone gets a number of reserved free pages based
473 proportionally on its size.
475 Some minimal amount of memory is needed to satisfy PF_MEMALLOC
476 allocations; if you set this to lower than 1024KB, your system will
477 become subtly broken, and prone to deadlock under high loads.
479 Setting this too high will OOM your machine instantly.
485 This is available only on NUMA kernels.
487 A percentage of the total pages in each zone. On Zone reclaim
488 (fallback from the local zone occurs) slabs will be reclaimed if more
489 than this percentage of pages in a zone are reclaimable slab pages.
490 This insures that the slab growth stays under control even in NUMA
491 systems that rarely perform global reclaim.
493 The default is 5 percent.
495 Note that slab reclaim is triggered in a per zone / node fashion.
496 The process of reclaiming slab memory is currently not node specific
503 This is available only on NUMA kernels.
505 This is a percentage of the total pages in each zone. Zone reclaim will
506 only occur if more than this percentage of pages are in a state that
507 zone_reclaim_mode allows to be reclaimed.
509 If zone_reclaim_mode has the value 4 OR'd, then the percentage is compared
510 against all file-backed unmapped pages including swapcache pages and tmpfs
511 files. Otherwise, only unmapped pages backed by normal files but not tmpfs
512 files and similar are considered.
514 The default is 1 percent.
520 This file indicates the amount of address space which a user process will
521 be restricted from mmapping. Since kernel null dereference bugs could
522 accidentally operate based on the information in the first couple of pages
523 of memory userspace processes should not be allowed to write to them. By
524 default this value is set to 0 and no protections will be enforced by the
525 security module. Setting this value to something like 64k will allow the
526 vast majority of applications to work correctly and provide defense in depth
527 against future potential kernel bugs.
533 This value can be used to select the number of bits to use to
534 determine the random offset to the base address of vma regions
535 resulting from mmap allocations on architectures which support
536 tuning address space randomization. This value will be bounded
537 by the architecture's minimum and maximum supported values.
539 This value can be changed after boot using the
540 /proc/sys/vm/mmap_rnd_bits tunable
546 This value can be used to select the number of bits to use to
547 determine the random offset to the base address of vma regions
548 resulting from mmap allocations for applications run in
549 compatibility mode on architectures which support tuning address
550 space randomization. This value will be bounded by the
551 architecture's minimum and maximum supported values.
553 This value can be changed after boot using the
554 /proc/sys/vm/mmap_rnd_compat_bits tunable
560 Change the minimum size of the hugepage pool.
562 See Documentation/admin-guide/mm/hugetlbpage.rst
565 hugetlb_optimize_vmemmap
566 ========================
568 This knob is not available when memory_hotplug.memmap_on_memory (kernel parameter)
569 is configured or the size of 'struct page' (a structure defined in
570 include/linux/mm_types.h) is not power of two (an unusual system config could
573 Enable (set to 1) or disable (set to 0) the feature of optimizing vmemmap pages
574 associated with each HugeTLB page.
576 Once enabled, the vmemmap pages of subsequent allocation of HugeTLB pages from
577 buddy allocator will be optimized (7 pages per 2MB HugeTLB page and 4095 pages
578 per 1GB HugeTLB page), whereas already allocated HugeTLB pages will not be
579 optimized. When those optimized HugeTLB pages are freed from the HugeTLB pool
580 to the buddy allocator, the vmemmap pages representing that range needs to be
581 remapped again and the vmemmap pages discarded earlier need to be rellocated
582 again. If your use case is that HugeTLB pages are allocated 'on the fly' (e.g.
583 never explicitly allocating HugeTLB pages with 'nr_hugepages' but only set
584 'nr_overcommit_hugepages', those overcommitted HugeTLB pages are allocated 'on
585 the fly') instead of being pulled from the HugeTLB pool, you should weigh the
586 benefits of memory savings against the more overhead (~2x slower than before)
587 of allocation or freeing HugeTLB pages between the HugeTLB pool and the buddy
588 allocator. Another behavior to note is that if the system is under heavy memory
589 pressure, it could prevent the user from freeing HugeTLB pages from the HugeTLB
590 pool to the buddy allocator since the allocation of vmemmap pages could be
591 failed, you have to retry later if your system encounter this situation.
593 Once disabled, the vmemmap pages of subsequent allocation of HugeTLB pages from
594 buddy allocator will not be optimized meaning the extra overhead at allocation
595 time from buddy allocator disappears, whereas already optimized HugeTLB pages
596 will not be affected. If you want to make sure there are no optimized HugeTLB
597 pages, you can set "nr_hugepages" to 0 first and then disable this. Note that
598 writing 0 to nr_hugepages will make any "in use" HugeTLB pages become surplus
599 pages. So, those surplus pages are still optimized until they are no longer
600 in use. You would need to wait for those surplus pages to be released before
601 there are no optimized pages in the system.
604 nr_hugepages_mempolicy
605 ======================
607 Change the size of the hugepage pool at run-time on a specific
610 See Documentation/admin-guide/mm/hugetlbpage.rst
613 nr_overcommit_hugepages
614 =======================
616 Change the maximum size of the hugepage pool. The maximum is
617 nr_hugepages + nr_overcommit_hugepages.
619 See Documentation/admin-guide/mm/hugetlbpage.rst
625 This is available only on NOMMU kernels.
627 This value adjusts the excess page trimming behaviour of power-of-2 aligned
628 NOMMU mmap allocations.
630 A value of 0 disables trimming of allocations entirely, while a value of 1
631 trims excess pages aggressively. Any value >= 1 acts as the watermark where
632 trimming of allocations is initiated.
634 The default value is 1.
636 See Documentation/admin-guide/mm/nommu-mmap.rst for more information.
642 This sysctl is only for NUMA and it is deprecated. Anything but
643 Node order will fail!
645 'where the memory is allocated from' is controlled by zonelists.
647 (This documentation ignores ZONE_HIGHMEM/ZONE_DMA32 for simple explanation.
648 you may be able to read ZONE_DMA as ZONE_DMA32...)
650 In non-NUMA case, a zonelist for GFP_KERNEL is ordered as following.
651 ZONE_NORMAL -> ZONE_DMA
652 This means that a memory allocation request for GFP_KERNEL will
653 get memory from ZONE_DMA only when ZONE_NORMAL is not available.
655 In NUMA case, you can think of following 2 types of order.
656 Assume 2 node NUMA and below is zonelist of Node(0)'s GFP_KERNEL::
658 (A) Node(0) ZONE_NORMAL -> Node(0) ZONE_DMA -> Node(1) ZONE_NORMAL
659 (B) Node(0) ZONE_NORMAL -> Node(1) ZONE_NORMAL -> Node(0) ZONE_DMA.
661 Type(A) offers the best locality for processes on Node(0), but ZONE_DMA
662 will be used before ZONE_NORMAL exhaustion. This increases possibility of
663 out-of-memory(OOM) of ZONE_DMA because ZONE_DMA is tend to be small.
665 Type(B) cannot offer the best locality but is more robust against OOM of
668 Type(A) is called as "Node" order. Type (B) is "Zone" order.
670 "Node order" orders the zonelists by node, then by zone within each node.
671 Specify "[Nn]ode" for node order
673 "Zone Order" orders the zonelists by zone type, then by node within each
674 zone. Specify "[Zz]one" for zone order.
676 Specify "[Dd]efault" to request automatic configuration.
678 On 32-bit, the Normal zone needs to be preserved for allocations accessible
679 by the kernel, so "zone" order will be selected.
681 On 64-bit, devices that require DMA32/DMA are relatively rare, so "node"
682 order will be selected.
684 Default order is recommended unless this is causing problems for your
691 Enables a system-wide task dump (excluding kernel threads) to be produced
692 when the kernel performs an OOM-killing and includes such information as
693 pid, uid, tgid, vm size, rss, pgtables_bytes, swapents, oom_score_adj
694 score, and name. This is helpful to determine why the OOM killer was
695 invoked, to identify the rogue task that caused it, and to determine why
696 the OOM killer chose the task it did to kill.
698 If this is set to zero, this information is suppressed. On very
699 large systems with thousands of tasks it may not be feasible to dump
700 the memory state information for each one. Such systems should not
701 be forced to incur a performance penalty in OOM conditions when the
702 information may not be desired.
704 If this is set to non-zero, this information is shown whenever the
705 OOM killer actually kills a memory-hogging task.
707 The default value is 1 (enabled).
710 oom_kill_allocating_task
711 ========================
713 This enables or disables killing the OOM-triggering task in
714 out-of-memory situations.
716 If this is set to zero, the OOM killer will scan through the entire
717 tasklist and select a task based on heuristics to kill. This normally
718 selects a rogue memory-hogging task that frees up a large amount of
721 If this is set to non-zero, the OOM killer simply kills the task that
722 triggered the out-of-memory condition. This avoids the expensive
725 If panic_on_oom is selected, it takes precedence over whatever value
726 is used in oom_kill_allocating_task.
728 The default value is 0.
734 When overcommit_memory is set to 2, the committed address space is not
735 permitted to exceed swap plus this amount of physical RAM. See below.
737 Note: overcommit_kbytes is the counterpart of overcommit_ratio. Only one
738 of them may be specified at a time. Setting one disables the other (which
739 then appears as 0 when read).
745 This value contains a flag that enables memory overcommitment.
747 When this flag is 0, the kernel attempts to estimate the amount
748 of free memory left when userspace requests more memory.
750 When this flag is 1, the kernel pretends there is always enough
751 memory until it actually runs out.
753 When this flag is 2, the kernel uses a "never overcommit"
754 policy that attempts to prevent any overcommit of memory.
755 Note that user_reserve_kbytes affects this policy.
757 This feature can be very useful because there are a lot of
758 programs that malloc() huge amounts of memory "just-in-case"
759 and don't use much of it.
761 The default value is 0.
763 See Documentation/vm/overcommit-accounting.rst and
764 mm/util.c::__vm_enough_memory() for more information.
770 When overcommit_memory is set to 2, the committed address
771 space is not permitted to exceed swap plus this percentage
772 of physical RAM. See above.
778 page-cluster controls the number of pages up to which consecutive pages
779 are read in from swap in a single attempt. This is the swap counterpart
780 to page cache readahead.
781 The mentioned consecutivity is not in terms of virtual/physical addresses,
782 but consecutive on swap space - that means they were swapped out together.
784 It is a logarithmic value - setting it to zero means "1 page", setting
785 it to 1 means "2 pages", setting it to 2 means "4 pages", etc.
786 Zero disables swap readahead completely.
788 The default value is three (eight pages at a time). There may be some
789 small benefits in tuning this to a different value if your workload is
792 Lower values mean lower latencies for initial faults, but at the same time
793 extra faults and I/O delays for following faults if they would have been part of
794 that consecutive pages readahead would have brought in.
800 This value determines the number of times that the page lock can be
801 stolen from under a waiter. After the lock is stolen the number of times
802 specified in this file (default is 5), the "fair lock handoff" semantics
803 will apply, and the waiter will only be awakened if the lock can be taken.
808 This enables or disables panic on out-of-memory feature.
810 If this is set to 0, the kernel will kill some rogue process,
811 called oom_killer. Usually, oom_killer can kill rogue processes and
814 If this is set to 1, the kernel panics when out-of-memory happens.
815 However, if a process limits using nodes by mempolicy/cpusets,
816 and those nodes become memory exhaustion status, one process
817 may be killed by oom-killer. No panic occurs in this case.
818 Because other nodes' memory may be free. This means system total status
819 may be not fatal yet.
821 If this is set to 2, the kernel panics compulsorily even on the
822 above-mentioned. Even oom happens under memory cgroup, the whole
825 The default value is 0.
827 1 and 2 are for failover of clustering. Please select either
828 according to your policy of failover.
830 panic_on_oom=2+kdump gives you very strong tool to investigate
831 why oom happens. You can get snapshot.
834 percpu_pagelist_high_fraction
835 =============================
837 This is the fraction of pages in each zone that are can be stored to
838 per-cpu page lists. It is an upper boundary that is divided depending
839 on the number of online CPUs. The min value for this is 8 which means
840 that we do not allow more than 1/8th of pages in each zone to be stored
841 on per-cpu page lists. This entry only changes the value of hot per-cpu
842 page lists. A user can specify a number like 100 to allocate 1/100th of
843 each zone between per-cpu lists.
845 The batch value of each per-cpu page list remains the same regardless of
846 the value of the high fraction so allocation latencies are unaffected.
848 The initial value is zero. Kernel uses this value to set the high pcp->high
849 mark based on the low watermark for the zone and the number of local
850 online CPUs. If the user writes '0' to this sysctl, it will revert to
851 this default behavior.
857 The time interval between which vm statistics are updated. The default
864 Any read or write (by root only) flushes all the per-cpu vm statistics
865 into their global totals, for more accurate reports when testing
866 e.g. cat /proc/sys/vm/stat_refresh /proc/meminfo
868 As a side-effect, it also checks for negative totals (elsewhere reported
869 as 0) and "fails" with EINVAL if any are found, with a warning in dmesg.
870 (At time of writing, a few stats are known sometimes to be found negative,
871 with no ill effects: errors and warnings on these stats are suppressed.)
877 This interface allows runtime configuration of numa statistics.
879 When page allocation performance becomes a bottleneck and you can tolerate
880 some possible tool breakage and decreased numa counter precision, you can
883 echo 0 > /proc/sys/vm/numa_stat
885 When page allocation performance is not a bottleneck and you want all
886 tooling to work, you can do::
888 echo 1 > /proc/sys/vm/numa_stat
894 This control is used to define the rough relative IO cost of swapping
895 and filesystem paging, as a value between 0 and 200. At 100, the VM
896 assumes equal IO cost and will thus apply memory pressure to the page
897 cache and swap-backed pages equally; lower values signify more
898 expensive swap IO, higher values indicates cheaper.
900 Keep in mind that filesystem IO patterns under memory pressure tend to
901 be more efficient than swap's random IO. An optimal value will require
902 experimentation and will also be workload-dependent.
904 The default value is 60.
906 For in-memory swap, like zram or zswap, as well as hybrid setups that
907 have swap on faster devices than the filesystem, values beyond 100 can
908 be considered. For example, if the random IO against the swap device
909 is on average 2x faster than IO from the filesystem, swappiness should
910 be 133 (x + 2x = 200, 2x = 133.33).
912 At 0, the kernel will not initiate swap until the amount of free and
913 file-backed pages is less than the high watermark in a zone.
916 unprivileged_userfaultfd
917 ========================
919 This flag controls the mode in which unprivileged users can use the
920 userfaultfd system calls. Set this to 0 to restrict unprivileged users
921 to handle page faults in user mode only. In this case, users without
922 SYS_CAP_PTRACE must pass UFFD_USER_MODE_ONLY in order for userfaultfd to
923 succeed. Prohibiting use of userfaultfd for handling faults from kernel
924 mode may make certain vulnerabilities more difficult to exploit.
926 Set this to 1 to allow unprivileged users to use the userfaultfd system
927 calls without any restrictions.
929 The default value is 0.
935 When overcommit_memory is set to 2, "never overcommit" mode, reserve
936 min(3% of current process size, user_reserve_kbytes) of free memory.
937 This is intended to prevent a user from starting a single memory hogging
938 process, such that they cannot recover (kill the hog).
940 user_reserve_kbytes defaults to min(3% of the current process size, 128MB).
942 If this is reduced to zero, then the user will be allowed to allocate
943 all free memory with a single process, minus admin_reserve_kbytes.
944 Any subsequent attempts to execute a command will result in
945 "fork: Cannot allocate memory".
947 Changing this takes effect whenever an application requests memory.
953 This percentage value controls the tendency of the kernel to reclaim
954 the memory which is used for caching of directory and inode objects.
956 At the default value of vfs_cache_pressure=100 the kernel will attempt to
957 reclaim dentries and inodes at a "fair" rate with respect to pagecache and
958 swapcache reclaim. Decreasing vfs_cache_pressure causes the kernel to prefer
959 to retain dentry and inode caches. When vfs_cache_pressure=0, the kernel will
960 never reclaim dentries and inodes due to memory pressure and this can easily
961 lead to out-of-memory conditions. Increasing vfs_cache_pressure beyond 100
962 causes the kernel to prefer to reclaim dentries and inodes.
964 Increasing vfs_cache_pressure significantly beyond 100 may have negative
965 performance impact. Reclaim code needs to take various locks to find freeable
966 directory and inode objects. With vfs_cache_pressure=1000, it will look for
967 ten times more freeable objects than there are.
970 watermark_boost_factor
971 ======================
973 This factor controls the level of reclaim when memory is being fragmented.
974 It defines the percentage of the high watermark of a zone that will be
975 reclaimed if pages of different mobility are being mixed within pageblocks.
976 The intent is that compaction has less work to do in the future and to
977 increase the success rate of future high-order allocations such as SLUB
978 allocations, THP and hugetlbfs pages.
980 To make it sensible with respect to the watermark_scale_factor
981 parameter, the unit is in fractions of 10,000. The default value of
982 15,000 means that up to 150% of the high watermark will be reclaimed in the
983 event of a pageblock being mixed due to fragmentation. The level of reclaim
984 is determined by the number of fragmentation events that occurred in the
985 recent past. If this value is smaller than a pageblock then a pageblocks
986 worth of pages will be reclaimed (e.g. 2MB on 64-bit x86). A boost factor
987 of 0 will disable the feature.
990 watermark_scale_factor
991 ======================
993 This factor controls the aggressiveness of kswapd. It defines the
994 amount of memory left in a node/system before kswapd is woken up and
995 how much memory needs to be free before kswapd goes back to sleep.
997 The unit is in fractions of 10,000. The default value of 10 means the
998 distances between watermarks are 0.1% of the available memory in the
999 node/system. The maximum value is 3000, or 30% of memory.
1001 A high rate of threads entering direct reclaim (allocstall) or kswapd
1002 going to sleep prematurely (kswapd_low_wmark_hit_quickly) can indicate
1003 that the number of free pages kswapd maintains for latency reasons is
1004 too small for the allocation bursts occurring in the system. This knob
1005 can then be used to tune kswapd aggressiveness accordingly.
1011 Zone_reclaim_mode allows someone to set more or less aggressive approaches to
1012 reclaim memory when a zone runs out of memory. If it is set to zero then no
1013 zone reclaim occurs. Allocations will be satisfied from other zones / nodes
1016 This is value OR'ed together of
1018 = ===================================
1020 2 Zone reclaim writes dirty pages out
1021 4 Zone reclaim swaps pages
1022 = ===================================
1024 zone_reclaim_mode is disabled by default. For file servers or workloads
1025 that benefit from having their data cached, zone_reclaim_mode should be
1026 left disabled as the caching effect is likely to be more important than
1029 Consider enabling one or more zone_reclaim mode bits if it's known that the
1030 workload is partitioned such that each partition fits within a NUMA node
1031 and that accessing remote memory would cause a measurable performance
1032 reduction. The page allocator will take additional actions before
1033 allocating off node pages.
1035 Allowing zone reclaim to write out pages stops processes that are
1036 writing large amounts of data from dirtying pages on other nodes. Zone
1037 reclaim will write out dirty pages if a zone fills up and so effectively
1038 throttle the process. This may decrease the performance of a single process
1039 since it cannot use all of system memory to buffer the outgoing writes
1040 anymore but it preserve the memory on other nodes so that the performance
1041 of other processes running on other nodes will not be affected.
1043 Allowing regular swap effectively restricts allocations to the local
1044 node unless explicitly overridden by memory policies or cpuset