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- compact_unevictable_allowed 25- dirty_background_bytes 26- dirty_background_ratio 27- dirty_bytes 28- dirty_expire_centisecs 29- dirty_ratio 30- dirty_writeback_centisecs 31- drop_caches 32- extfrag_threshold 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_kbytes 54- overcommit_memory 55- overcommit_ratio 56- page-cluster 57- panic_on_oom 58- percpu_pagelist_fraction 59- stat_interval 60- stat_refresh 61- swappiness 62- user_reserve_kbytes 63- vfs_cache_pressure 64- watermark_scale_factor 65- zone_reclaim_mode 66 67============================================================== 68 69admin_reserve_kbytes 70 71The amount of free memory in the system that should be reserved for users 72with the capability cap_sys_admin. 73 74admin_reserve_kbytes defaults to min(3% of free pages, 8MB) 75 76That should provide enough for the admin to log in and kill a process, 77if necessary, under the default overcommit 'guess' mode. 78 79Systems running under overcommit 'never' should increase this to account 80for the full Virtual Memory Size of programs used to recover. Otherwise, 81root may not be able to log in to recover the system. 82 83How do you calculate a minimum useful reserve? 84 85sshd or login + bash (or some other shell) + top (or ps, kill, etc.) 86 87For overcommit 'guess', we can sum resident set sizes (RSS). 88On x86_64 this is about 8MB. 89 90For overcommit 'never', we can take the max of their virtual sizes (VSZ) 91and add the sum of their RSS. 92On x86_64 this is about 128MB. 93 94Changing this takes effect whenever an application requests memory. 95 96============================================================== 97 98block_dump 99 100block_dump enables block I/O debugging when set to a nonzero value. More 101information on block I/O debugging is in Documentation/laptops/laptop-mode.txt. 102 103============================================================== 104 105compact_memory 106 107Available only when CONFIG_COMPACTION is set. When 1 is written to the file, 108all zones are compacted such that free memory is available in contiguous 109blocks where possible. This can be important for example in the allocation of 110huge pages although processes will also directly compact memory as required. 111 112============================================================== 113 114compact_unevictable_allowed 115 116Available only when CONFIG_COMPACTION is set. When set to 1, compaction is 117allowed to examine the unevictable lru (mlocked pages) for pages to compact. 118This should be used on systems where stalls for minor page faults are an 119acceptable trade for large contiguous free memory. Set to 0 to prevent 120compaction from moving pages that are unevictable. Default value is 1. 121 122============================================================== 123 124dirty_background_bytes 125 126Contains the amount of dirty memory at which the background kernel 127flusher threads will start writeback. 128 129Note: dirty_background_bytes is the counterpart of dirty_background_ratio. Only 130one of them may be specified at a time. When one sysctl is written it is 131immediately taken into account to evaluate the dirty memory limits and the 132other appears as 0 when read. 133 134============================================================== 135 136dirty_background_ratio 137 138Contains, as a percentage of total available memory that contains free pages 139and reclaimable pages, the number of pages at which the background kernel 140flusher threads will start writing out dirty data. 141 142The total available memory is not equal to total system memory. 143 144============================================================== 145 146dirty_bytes 147 148Contains the amount of dirty memory at which a process generating disk writes 149will itself start writeback. 150 151Note: dirty_bytes is the counterpart of dirty_ratio. Only one of them may be 152specified at a time. When one sysctl is written it is immediately taken into 153account to evaluate the dirty memory limits and the other appears as 0 when 154read. 155 156Note: the minimum value allowed for dirty_bytes is two pages (in bytes); any 157value lower than this limit will be ignored and the old configuration will be 158retained. 159 160============================================================== 161 162dirty_expire_centisecs 163 164This tunable is used to define when dirty data is old enough to be eligible 165for writeout by the kernel flusher threads. It is expressed in 100'ths 166of a second. Data which has been dirty in-memory for longer than this 167interval will be written out next time a flusher thread wakes up. 168 169============================================================== 170 171dirty_ratio 172 173Contains, as a percentage of total available memory that contains free pages 174and reclaimable pages, the number of pages at which a process which is 175generating disk writes will itself start writing out dirty data. 176 177The total available memory is not equal to total system memory. 178 179============================================================== 180 181dirty_writeback_centisecs 182 183The kernel flusher threads will periodically wake up and write `old' data 184out to disk. This tunable expresses the interval between those wakeups, in 185100'ths of a second. 186 187Setting this to zero disables periodic writeback altogether. 188 189============================================================== 190 191drop_caches 192 193Writing to this will cause the kernel to drop clean caches, as well as 194reclaimable slab objects like dentries and inodes. Once dropped, their 195memory becomes free. 196 197To free pagecache: 198 echo 1 > /proc/sys/vm/drop_caches 199To free reclaimable slab objects (includes dentries and inodes): 200 echo 2 > /proc/sys/vm/drop_caches 201To free slab objects and pagecache: 202 echo 3 > /proc/sys/vm/drop_caches 203 204This is a non-destructive operation and will not free any dirty objects. 205To increase the number of objects freed by this operation, the user may run 206`sync' prior to writing to /proc/sys/vm/drop_caches. This will minimize the 207number of dirty objects on the system and create more candidates to be 208dropped. 209 210This file is not a means to control the growth of the various kernel caches 211(inodes, dentries, pagecache, etc...) These objects are automatically 212reclaimed by the kernel when memory is needed elsewhere on the system. 213 214Use of this file can cause performance problems. Since it discards cached 215objects, it may cost a significant amount of I/O and CPU to recreate the 216dropped objects, especially if they were under heavy use. Because of this, 217use outside of a testing or debugging environment is not recommended. 218 219You may see informational messages in your kernel log when this file is 220used: 221 222 cat (1234): drop_caches: 3 223 224These are informational only. They do not mean that anything is wrong 225with your system. To disable them, echo 4 (bit 3) into drop_caches. 226 227============================================================== 228 229extfrag_threshold 230 231This parameter affects whether the kernel will compact memory or direct 232reclaim to satisfy a high-order allocation. The extfrag/extfrag_index file in 233debugfs shows what the fragmentation index for each order is in each zone in 234the system. Values tending towards 0 imply allocations would fail due to lack 235of memory, values towards 1000 imply failures are due to fragmentation and -1 236implies that the allocation will succeed as long as watermarks are met. 237 238The kernel will not compact memory in a zone if the 239fragmentation index is <= extfrag_threshold. The default value is 500. 240 241============================================================== 242 243hugepages_treat_as_movable 244 245This parameter controls whether we can allocate hugepages from ZONE_MOVABLE 246or not. If set to non-zero, hugepages can be allocated from ZONE_MOVABLE. 247ZONE_MOVABLE is created when kernel boot parameter kernelcore= is specified, 248so this parameter has no effect if used without kernelcore=. 249 250Hugepage migration is now available in some situations which depend on the 251architecture and/or the hugepage size. If a hugepage supports migration, 252allocation from ZONE_MOVABLE is always enabled for the hugepage regardless 253of the value of this parameter. 254IOW, this parameter affects only non-migratable hugepages. 255 256Assuming that hugepages are not migratable in your system, one usecase of 257this parameter is that users can make hugepage pool more extensible by 258enabling the allocation from ZONE_MOVABLE. This is because on ZONE_MOVABLE 259page reclaim/migration/compaction work more and you can get contiguous 260memory more likely. Note that using ZONE_MOVABLE for non-migratable 261hugepages can do harm to other features like memory hotremove (because 262memory hotremove expects that memory blocks on ZONE_MOVABLE are always 263removable,) so it's a trade-off responsible for the users. 264 265============================================================== 266 267hugetlb_shm_group 268 269hugetlb_shm_group contains group id that is allowed to create SysV 270shared memory segment using hugetlb page. 271 272============================================================== 273 274laptop_mode 275 276laptop_mode is a knob that controls "laptop mode". All the things that are 277controlled by this knob are discussed in Documentation/laptops/laptop-mode.txt. 278 279============================================================== 280 281legacy_va_layout 282 283If non-zero, this sysctl disables the new 32-bit mmap layout - the kernel 284will use the legacy (2.4) layout for all processes. 285 286============================================================== 287 288lowmem_reserve_ratio 289 290For some specialised workloads on highmem machines it is dangerous for 291the kernel to allow process memory to be allocated from the "lowmem" 292zone. This is because that memory could then be pinned via the mlock() 293system call, or by unavailability of swapspace. 294 295And on large highmem machines this lack of reclaimable lowmem memory 296can be fatal. 297 298So the Linux page allocator has a mechanism which prevents allocations 299which _could_ use highmem from using too much lowmem. This means that 300a certain amount of lowmem is defended from the possibility of being 301captured into pinned user memory. 302 303(The same argument applies to the old 16 megabyte ISA DMA region. This 304mechanism will also defend that region from allocations which could use 305highmem or lowmem). 306 307The `lowmem_reserve_ratio' tunable determines how aggressive the kernel is 308in defending these lower zones. 309 310If you have a machine which uses highmem or ISA DMA and your 311applications are using mlock(), or if you are running with no swap then 312you probably should change the lowmem_reserve_ratio setting. 313 314The lowmem_reserve_ratio is an array. You can see them by reading this file. 315- 316% cat /proc/sys/vm/lowmem_reserve_ratio 317256 256 32 318- 319Note: # of this elements is one fewer than number of zones. Because the highest 320 zone's value is not necessary for following calculation. 321 322But, these values are not used directly. The kernel calculates # of protection 323pages for each zones from them. These are shown as array of protection pages 324in /proc/zoneinfo like followings. (This is an example of x86-64 box). 325Each zone has an array of protection pages like this. 326 327- 328Node 0, zone DMA 329 pages free 1355 330 min 3 331 low 3 332 high 4 333 : 334 : 335 numa_other 0 336 protection: (0, 2004, 2004, 2004) 337 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 338 pagesets 339 cpu: 0 pcp: 0 340 : 341- 342These protections are added to score to judge whether this zone should be used 343for page allocation or should be reclaimed. 344 345In this example, if normal pages (index=2) are required to this DMA zone and 346watermark[WMARK_HIGH] is used for watermark, the kernel judges this zone should 347not be used because pages_free(1355) is smaller than watermark + protection[2] 348(4 + 2004 = 2008). If this protection value is 0, this zone would be used for 349normal page requirement. If requirement is DMA zone(index=0), protection[0] 350(=0) is used. 351 352zone[i]'s protection[j] is calculated by following expression. 353 354(i < j): 355 zone[i]->protection[j] 356 = (total sums of managed_pages from zone[i+1] to zone[j] on the node) 357 / lowmem_reserve_ratio[i]; 358(i = j): 359 (should not be protected. = 0; 360(i > j): 361 (not necessary, but looks 0) 362 363The default values of lowmem_reserve_ratio[i] are 364 256 (if zone[i] means DMA or DMA32 zone) 365 32 (others). 366As above expression, they are reciprocal number of ratio. 367256 means 1/256. # of protection pages becomes about "0.39%" of total managed 368pages of higher zones on the node. 369 370If you would like to protect more pages, smaller values are effective. 371The minimum value is 1 (1/1 -> 100%). 372 373============================================================== 374 375max_map_count: 376 377This file contains the maximum number of memory map areas a process 378may have. Memory map areas are used as a side-effect of calling 379malloc, directly by mmap and mprotect, and also when loading shared 380libraries. 381 382While most applications need less than a thousand maps, certain 383programs, particularly malloc debuggers, may consume lots of them, 384e.g., up to one or two maps per allocation. 385 386The default value is 65536. 387 388============================================================= 389 390memory_failure_early_kill: 391 392Control how to kill processes when uncorrected memory error (typically 393a 2bit error in a memory module) is detected in the background by hardware 394that cannot be handled by the kernel. In some cases (like the page 395still having a valid copy on disk) the kernel will handle the failure 396transparently without affecting any applications. But if there is 397no other uptodate copy of the data it will kill to prevent any data 398corruptions from propagating. 399 4001: Kill all processes that have the corrupted and not reloadable page mapped 401as soon as the corruption is detected. Note this is not supported 402for a few types of pages, like kernel internally allocated data or 403the swap cache, but works for the majority of user pages. 404 4050: Only unmap the corrupted page from all processes and only kill a process 406who tries to access it. 407 408The kill is done using a catchable SIGBUS with BUS_MCEERR_AO, so processes can 409handle this if they want to. 410 411This is only active on architectures/platforms with advanced machine 412check handling and depends on the hardware capabilities. 413 414Applications can override this setting individually with the PR_MCE_KILL prctl 415 416============================================================== 417 418memory_failure_recovery 419 420Enable memory failure recovery (when supported by the platform) 421 4221: Attempt recovery. 423 4240: Always panic on a memory failure. 425 426============================================================== 427 428min_free_kbytes: 429 430This is used to force the Linux VM to keep a minimum number 431of kilobytes free. The VM uses this number to compute a 432watermark[WMARK_MIN] value for each lowmem zone in the system. 433Each lowmem zone gets a number of reserved free pages based 434proportionally on its size. 435 436Some minimal amount of memory is needed to satisfy PF_MEMALLOC 437allocations; if you set this to lower than 1024KB, your system will 438become subtly broken, and prone to deadlock under high loads. 439 440Setting this too high will OOM your machine instantly. 441 442============================================================= 443 444min_slab_ratio: 445 446This is available only on NUMA kernels. 447 448A percentage of the total pages in each zone. On Zone reclaim 449(fallback from the local zone occurs) slabs will be reclaimed if more 450than this percentage of pages in a zone are reclaimable slab pages. 451This insures that the slab growth stays under control even in NUMA 452systems that rarely perform global reclaim. 453 454The default is 5 percent. 455 456Note that slab reclaim is triggered in a per zone / node fashion. 457The process of reclaiming slab memory is currently not node specific 458and may not be fast. 459 460============================================================= 461 462min_unmapped_ratio: 463 464This is available only on NUMA kernels. 465 466This is a percentage of the total pages in each zone. Zone reclaim will 467only occur if more than this percentage of pages are in a state that 468zone_reclaim_mode allows to be reclaimed. 469 470If zone_reclaim_mode has the value 4 OR'd, then the percentage is compared 471against all file-backed unmapped pages including swapcache pages and tmpfs 472files. Otherwise, only unmapped pages backed by normal files but not tmpfs 473files and similar are considered. 474 475The default is 1 percent. 476 477============================================================== 478 479mmap_min_addr 480 481This file indicates the amount of address space which a user process will 482be restricted from mmapping. Since kernel null dereference bugs could 483accidentally operate based on the information in the first couple of pages 484of memory userspace processes should not be allowed to write to them. By 485default this value is set to 0 and no protections will be enforced by the 486security module. Setting this value to something like 64k will allow the 487vast majority of applications to work correctly and provide defense in depth 488against future potential kernel bugs. 489 490============================================================== 491 492mmap_rnd_bits: 493 494This value can be used to select the number of bits to use to 495determine the random offset to the base address of vma regions 496resulting from mmap allocations on architectures which support 497tuning address space randomization. This value will be bounded 498by the architecture's minimum and maximum supported values. 499 500This value can be changed after boot using the 501/proc/sys/vm/mmap_rnd_bits tunable 502 503============================================================== 504 505mmap_rnd_compat_bits: 506 507This value can be used to select the number of bits to use to 508determine the random offset to the base address of vma regions 509resulting from mmap allocations for applications run in 510compatibility mode on architectures which support tuning address 511space randomization. This value will be bounded by the 512architecture's minimum and maximum supported values. 513 514This value can be changed after boot using the 515/proc/sys/vm/mmap_rnd_compat_bits tunable 516 517============================================================== 518 519nr_hugepages 520 521Change the minimum size of the hugepage pool. 522 523See Documentation/vm/hugetlbpage.txt 524 525============================================================== 526 527nr_overcommit_hugepages 528 529Change the maximum size of the hugepage pool. The maximum is 530nr_hugepages + nr_overcommit_hugepages. 531 532See Documentation/vm/hugetlbpage.txt 533 534============================================================== 535 536nr_trim_pages 537 538This is available only on NOMMU kernels. 539 540This value adjusts the excess page trimming behaviour of power-of-2 aligned 541NOMMU mmap allocations. 542 543A value of 0 disables trimming of allocations entirely, while a value of 1 544trims excess pages aggressively. Any value >= 1 acts as the watermark where 545trimming of allocations is initiated. 546 547The default value is 1. 548 549See Documentation/nommu-mmap.txt for more information. 550 551============================================================== 552 553numa_zonelist_order 554 555This sysctl is only for NUMA. 556'where the memory is allocated from' is controlled by zonelists. 557(This documentation ignores ZONE_HIGHMEM/ZONE_DMA32 for simple explanation. 558 you may be able to read ZONE_DMA as ZONE_DMA32...) 559 560In non-NUMA case, a zonelist for GFP_KERNEL is ordered as following. 561ZONE_NORMAL -> ZONE_DMA 562This means that a memory allocation request for GFP_KERNEL will 563get memory from ZONE_DMA only when ZONE_NORMAL is not available. 564 565In NUMA case, you can think of following 2 types of order. 566Assume 2 node NUMA and below is zonelist of Node(0)'s GFP_KERNEL 567 568(A) Node(0) ZONE_NORMAL -> Node(0) ZONE_DMA -> Node(1) ZONE_NORMAL 569(B) Node(0) ZONE_NORMAL -> Node(1) ZONE_NORMAL -> Node(0) ZONE_DMA. 570 571Type(A) offers the best locality for processes on Node(0), but ZONE_DMA 572will be used before ZONE_NORMAL exhaustion. This increases possibility of 573out-of-memory(OOM) of ZONE_DMA because ZONE_DMA is tend to be small. 574 575Type(B) cannot offer the best locality but is more robust against OOM of 576the DMA zone. 577 578Type(A) is called as "Node" order. Type (B) is "Zone" order. 579 580"Node order" orders the zonelists by node, then by zone within each node. 581Specify "[Nn]ode" for node order 582 583"Zone Order" orders the zonelists by zone type, then by node within each 584zone. Specify "[Zz]one" for zone order. 585 586Specify "[Dd]efault" to request automatic configuration. 587 588On 32-bit, the Normal zone needs to be preserved for allocations accessible 589by the kernel, so "zone" order will be selected. 590 591On 64-bit, devices that require DMA32/DMA are relatively rare, so "node" 592order will be selected. 593 594Default order is recommended unless this is causing problems for your 595system/application. 596 597============================================================== 598 599oom_dump_tasks 600 601Enables a system-wide task dump (excluding kernel threads) to be produced 602when the kernel performs an OOM-killing and includes such information as 603pid, uid, tgid, vm size, rss, nr_ptes, nr_pmds, swapents, oom_score_adj 604score, and name. This is helpful to determine why the OOM killer was 605invoked, to identify the rogue task that caused it, and to determine why 606the OOM killer chose the task it did to kill. 607 608If this is set to zero, this information is suppressed. On very 609large systems with thousands of tasks it may not be feasible to dump 610the memory state information for each one. Such systems should not 611be forced to incur a performance penalty in OOM conditions when the 612information may not be desired. 613 614If this is set to non-zero, this information is shown whenever the 615OOM killer actually kills a memory-hogging task. 616 617The default value is 1 (enabled). 618 619============================================================== 620 621oom_kill_allocating_task 622 623This enables or disables killing the OOM-triggering task in 624out-of-memory situations. 625 626If this is set to zero, the OOM killer will scan through the entire 627tasklist and select a task based on heuristics to kill. This normally 628selects a rogue memory-hogging task that frees up a large amount of 629memory when killed. 630 631If this is set to non-zero, the OOM killer simply kills the task that 632triggered the out-of-memory condition. This avoids the expensive 633tasklist scan. 634 635If panic_on_oom is selected, it takes precedence over whatever value 636is used in oom_kill_allocating_task. 637 638The default value is 0. 639 640============================================================== 641 642overcommit_kbytes: 643 644When overcommit_memory is set to 2, the committed address space is not 645permitted to exceed swap plus this amount of physical RAM. See below. 646 647Note: overcommit_kbytes is the counterpart of overcommit_ratio. Only one 648of them may be specified at a time. Setting one disables the other (which 649then appears as 0 when read). 650 651============================================================== 652 653overcommit_memory: 654 655This value contains a flag that enables memory overcommitment. 656 657When this flag is 0, the kernel attempts to estimate the amount 658of free memory left when userspace requests more memory. 659 660When this flag is 1, the kernel pretends there is always enough 661memory until it actually runs out. 662 663When this flag is 2, the kernel uses a "never overcommit" 664policy that attempts to prevent any overcommit of memory. 665Note that user_reserve_kbytes affects this policy. 666 667This feature can be very useful because there are a lot of 668programs that malloc() huge amounts of memory "just-in-case" 669and don't use much of it. 670 671The default value is 0. 672 673See Documentation/vm/overcommit-accounting and 674mm/mmap.c::__vm_enough_memory() for more information. 675 676============================================================== 677 678overcommit_ratio: 679 680When overcommit_memory is set to 2, the committed address 681space is not permitted to exceed swap plus this percentage 682of physical RAM. See above. 683 684============================================================== 685 686page-cluster 687 688page-cluster controls the number of pages up to which consecutive pages 689are read in from swap in a single attempt. This is the swap counterpart 690to page cache readahead. 691The mentioned consecutivity is not in terms of virtual/physical addresses, 692but consecutive on swap space - that means they were swapped out together. 693 694It is a logarithmic value - setting it to zero means "1 page", setting 695it to 1 means "2 pages", setting it to 2 means "4 pages", etc. 696Zero disables swap readahead completely. 697 698The default value is three (eight pages at a time). There may be some 699small benefits in tuning this to a different value if your workload is 700swap-intensive. 701 702Lower values mean lower latencies for initial faults, but at the same time 703extra faults and I/O delays for following faults if they would have been part of 704that consecutive pages readahead would have brought in. 705 706============================================================= 707 708panic_on_oom 709 710This enables or disables panic on out-of-memory feature. 711 712If this is set to 0, the kernel will kill some rogue process, 713called oom_killer. Usually, oom_killer can kill rogue processes and 714system will survive. 715 716If this is set to 1, the kernel panics when out-of-memory happens. 717However, if a process limits using nodes by mempolicy/cpusets, 718and those nodes become memory exhaustion status, one process 719may be killed by oom-killer. No panic occurs in this case. 720Because other nodes' memory may be free. This means system total status 721may be not fatal yet. 722 723If this is set to 2, the kernel panics compulsorily even on the 724above-mentioned. Even oom happens under memory cgroup, the whole 725system panics. 726 727The default value is 0. 7281 and 2 are for failover of clustering. Please select either 729according to your policy of failover. 730panic_on_oom=2+kdump gives you very strong tool to investigate 731why oom happens. You can get snapshot. 732 733============================================================= 734 735percpu_pagelist_fraction 736 737This is the fraction of pages at most (high mark pcp->high) in each zone that 738are allocated for each per cpu page list. The min value for this is 8. It 739means that we don't allow more than 1/8th of pages in each zone to be 740allocated in any single per_cpu_pagelist. This entry only changes the value 741of hot per cpu pagelists. User can specify a number like 100 to allocate 7421/100th of each zone to each per cpu page list. 743 744The batch value of each per cpu pagelist is also updated as a result. It is 745set to pcp->high/4. The upper limit of batch is (PAGE_SHIFT * 8) 746 747The initial value is zero. Kernel does not use this value at boot time to set 748the high water marks for each per cpu page list. If the user writes '0' to this 749sysctl, it will revert to this default behavior. 750 751============================================================== 752 753stat_interval 754 755The time interval between which vm statistics are updated. The default 756is 1 second. 757 758============================================================== 759 760stat_refresh 761 762Any read or write (by root only) flushes all the per-cpu vm statistics 763into their global totals, for more accurate reports when testing 764e.g. cat /proc/sys/vm/stat_refresh /proc/meminfo 765 766As a side-effect, it also checks for negative totals (elsewhere reported 767as 0) and "fails" with EINVAL if any are found, with a warning in dmesg. 768(At time of writing, a few stats are known sometimes to be found negative, 769with no ill effects: errors and warnings on these stats are suppressed.) 770 771============================================================== 772 773swappiness 774 775This control is used to define how aggressive the kernel will swap 776memory pages. Higher values will increase agressiveness, lower values 777decrease the amount of swap. A value of 0 instructs the kernel not to 778initiate swap until the amount of free and file-backed pages is less 779than the high water mark in a zone. 780 781The default value is 60. 782 783============================================================== 784 785- user_reserve_kbytes 786 787When overcommit_memory is set to 2, "never overcommit" mode, reserve 788min(3% of current process size, user_reserve_kbytes) of free memory. 789This is intended to prevent a user from starting a single memory hogging 790process, such that they cannot recover (kill the hog). 791 792user_reserve_kbytes defaults to min(3% of the current process size, 128MB). 793 794If this is reduced to zero, then the user will be allowed to allocate 795all free memory with a single process, minus admin_reserve_kbytes. 796Any subsequent attempts to execute a command will result in 797"fork: Cannot allocate memory". 798 799Changing this takes effect whenever an application requests memory. 800 801============================================================== 802 803vfs_cache_pressure 804------------------ 805 806This percentage value controls the tendency of the kernel to reclaim 807the memory which is used for caching of directory and inode objects. 808 809At the default value of vfs_cache_pressure=100 the kernel will attempt to 810reclaim dentries and inodes at a "fair" rate with respect to pagecache and 811swapcache reclaim. Decreasing vfs_cache_pressure causes the kernel to prefer 812to retain dentry and inode caches. When vfs_cache_pressure=0, the kernel will 813never reclaim dentries and inodes due to memory pressure and this can easily 814lead to out-of-memory conditions. Increasing vfs_cache_pressure beyond 100 815causes the kernel to prefer to reclaim dentries and inodes. 816 817Increasing vfs_cache_pressure significantly beyond 100 may have negative 818performance impact. Reclaim code needs to take various locks to find freeable 819directory and inode objects. With vfs_cache_pressure=1000, it will look for 820ten times more freeable objects than there are. 821 822============================================================= 823 824watermark_scale_factor: 825 826This factor controls the aggressiveness of kswapd. It defines the 827amount of memory left in a node/system before kswapd is woken up and 828how much memory needs to be free before kswapd goes back to sleep. 829 830The unit is in fractions of 10,000. The default value of 10 means the 831distances between watermarks are 0.1% of the available memory in the 832node/system. The maximum value is 1000, or 10% of memory. 833 834A high rate of threads entering direct reclaim (allocstall) or kswapd 835going to sleep prematurely (kswapd_low_wmark_hit_quickly) can indicate 836that the number of free pages kswapd maintains for latency reasons is 837too small for the allocation bursts occurring in the system. This knob 838can then be used to tune kswapd aggressiveness accordingly. 839 840============================================================== 841 842zone_reclaim_mode: 843 844Zone_reclaim_mode allows someone to set more or less aggressive approaches to 845reclaim memory when a zone runs out of memory. If it is set to zero then no 846zone reclaim occurs. Allocations will be satisfied from other zones / nodes 847in the system. 848 849This is value ORed together of 850 8511 = Zone reclaim on 8522 = Zone reclaim writes dirty pages out 8534 = Zone reclaim swaps pages 854 855zone_reclaim_mode is disabled by default. For file servers or workloads 856that benefit from having their data cached, zone_reclaim_mode should be 857left disabled as the caching effect is likely to be more important than 858data locality. 859 860zone_reclaim may be enabled if it's known that the workload is partitioned 861such that each partition fits within a NUMA node and that accessing remote 862memory would cause a measurable performance reduction. The page allocator 863will then reclaim easily reusable pages (those page cache pages that are 864currently not used) before allocating off node pages. 865 866Allowing zone reclaim to write out pages stops processes that are 867writing large amounts of data from dirtying pages on other nodes. Zone 868reclaim will write out dirty pages if a zone fills up and so effectively 869throttle the process. This may decrease the performance of a single process 870since it cannot use all of system memory to buffer the outgoing writes 871anymore but it preserve the memory on other nodes so that the performance 872of other processes running on other nodes will not be affected. 873 874Allowing regular swap effectively restricts allocations to the local 875node unless explicitly overridden by memory policies or cpuset 876configurations. 877 878============ End of Document ================================= 879