Based on kernel version 4.15. Page generated on 2018-01-29 10:01 EST.
1 Documentation for /proc/sys/vm/* kernel version 2.6.29 2 (c) 1998, 1999, Rik van Riel <firstname.lastname@example.org> 3 (c) 2008 Peter W. Morreale <email@example.com> 4 5 For general info and legal blurb, please look in README. 6 7 ============================================================== 8 9 This file contains the documentation for the sysctl files in 10 /proc/sys/vm and is valid for Linux kernel version 2.6.29. 11 12 The files in this directory can be used to tune the operation 13 of the virtual memory (VM) subsystem of the Linux kernel and 14 the writeout of dirty data to disk. 15 16 Default values and initialization routines for most of these 17 files can be found in mm/swap.c. 18 19 Currently, 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 - numa_stat 62 - swappiness 63 - user_reserve_kbytes 64 - vfs_cache_pressure 65 - watermark_scale_factor 66 - zone_reclaim_mode 67 68 ============================================================== 69 70 admin_reserve_kbytes 71 72 The amount of free memory in the system that should be reserved for users 73 with the capability cap_sys_admin. 74 75 admin_reserve_kbytes defaults to min(3% of free pages, 8MB) 76 77 That should provide enough for the admin to log in and kill a process, 78 if necessary, under the default overcommit 'guess' mode. 79 80 Systems running under overcommit 'never' should increase this to account 81 for the full Virtual Memory Size of programs used to recover. Otherwise, 82 root may not be able to log in to recover the system. 83 84 How do you calculate a minimum useful reserve? 85 86 sshd or login + bash (or some other shell) + top (or ps, kill, etc.) 87 88 For overcommit 'guess', we can sum resident set sizes (RSS). 89 On x86_64 this is about 8MB. 90 91 For overcommit 'never', we can take the max of their virtual sizes (VSZ) 92 and add the sum of their RSS. 93 On x86_64 this is about 128MB. 94 95 Changing this takes effect whenever an application requests memory. 96 97 ============================================================== 98 99 block_dump 100 101 block_dump enables block I/O debugging when set to a nonzero value. More 102 information on block I/O debugging is in Documentation/laptops/laptop-mode.txt. 103 104 ============================================================== 105 106 compact_memory 107 108 Available only when CONFIG_COMPACTION is set. When 1 is written to the file, 109 all zones are compacted such that free memory is available in contiguous 110 blocks where possible. This can be important for example in the allocation of 111 huge pages although processes will also directly compact memory as required. 112 113 ============================================================== 114 115 compact_unevictable_allowed 116 117 Available only when CONFIG_COMPACTION is set. When set to 1, compaction is 118 allowed to examine the unevictable lru (mlocked pages) for pages to compact. 119 This should be used on systems where stalls for minor page faults are an 120 acceptable trade for large contiguous free memory. Set to 0 to prevent 121 compaction from moving pages that are unevictable. Default value is 1. 122 123 ============================================================== 124 125 dirty_background_bytes 126 127 Contains the amount of dirty memory at which the background kernel 128 flusher threads will start writeback. 129 130 Note: dirty_background_bytes is the counterpart of dirty_background_ratio. Only 131 one of them may be specified at a time. When one sysctl is written it is 132 immediately taken into account to evaluate the dirty memory limits and the 133 other appears as 0 when read. 134 135 ============================================================== 136 137 dirty_background_ratio 138 139 Contains, as a percentage of total available memory that contains free pages 140 and reclaimable pages, the number of pages at which the background kernel 141 flusher threads will start writing out dirty data. 142 143 The total available memory is not equal to total system memory. 144 145 ============================================================== 146 147 dirty_bytes 148 149 Contains the amount of dirty memory at which a process generating disk writes 150 will itself start writeback. 151 152 Note: dirty_bytes is the counterpart of dirty_ratio. Only one of them may be 153 specified at a time. When one sysctl is written it is immediately taken into 154 account to evaluate the dirty memory limits and the other appears as 0 when 155 read. 156 157 Note: the minimum value allowed for dirty_bytes is two pages (in bytes); any 158 value lower than this limit will be ignored and the old configuration will be 159 retained. 160 161 ============================================================== 162 163 dirty_expire_centisecs 164 165 This tunable is used to define when dirty data is old enough to be eligible 166 for writeout by the kernel flusher threads. It is expressed in 100'ths 167 of a second. Data which has been dirty in-memory for longer than this 168 interval will be written out next time a flusher thread wakes up. 169 170 ============================================================== 171 172 dirty_ratio 173 174 Contains, as a percentage of total available memory that contains free pages 175 and reclaimable pages, the number of pages at which a process which is 176 generating disk writes will itself start writing out dirty data. 177 178 The total available memory is not equal to total system memory. 179 180 ============================================================== 181 182 dirty_writeback_centisecs 183 184 The kernel flusher threads will periodically wake up and write `old' data 185 out to disk. This tunable expresses the interval between those wakeups, in 186 100'ths of a second. 187 188 Setting this to zero disables periodic writeback altogether. 189 190 ============================================================== 191 192 drop_caches 193 194 Writing to this will cause the kernel to drop clean caches, as well as 195 reclaimable slab objects like dentries and inodes. Once dropped, their 196 memory becomes free. 197 198 To free pagecache: 199 echo 1 > /proc/sys/vm/drop_caches 200 To free reclaimable slab objects (includes dentries and inodes): 201 echo 2 > /proc/sys/vm/drop_caches 202 To free slab objects and pagecache: 203 echo 3 > /proc/sys/vm/drop_caches 204 205 This is a non-destructive operation and will not free any dirty objects. 206 To increase the number of objects freed by this operation, the user may run 207 `sync' prior to writing to /proc/sys/vm/drop_caches. This will minimize the 208 number of dirty objects on the system and create more candidates to be 209 dropped. 210 211 This file is not a means to control the growth of the various kernel caches 212 (inodes, dentries, pagecache, etc...) These objects are automatically 213 reclaimed by the kernel when memory is needed elsewhere on the system. 214 215 Use of this file can cause performance problems. Since it discards cached 216 objects, it may cost a significant amount of I/O and CPU to recreate the 217 dropped objects, especially if they were under heavy use. Because of this, 218 use outside of a testing or debugging environment is not recommended. 219 220 You may see informational messages in your kernel log when this file is 221 used: 222 223 cat (1234): drop_caches: 3 224 225 These are informational only. They do not mean that anything is wrong 226 with your system. To disable them, echo 4 (bit 3) into drop_caches. 227 228 ============================================================== 229 230 extfrag_threshold 231 232 This parameter affects whether the kernel will compact memory or direct 233 reclaim to satisfy a high-order allocation. The extfrag/extfrag_index file in 234 debugfs shows what the fragmentation index for each order is in each zone in 235 the system. Values tending towards 0 imply allocations would fail due to lack 236 of memory, values towards 1000 imply failures are due to fragmentation and -1 237 implies that the allocation will succeed as long as watermarks are met. 238 239 The kernel will not compact memory in a zone if the 240 fragmentation index is <= extfrag_threshold. The default value is 500. 241 242 ============================================================== 243 244 highmem_is_dirtyable 245 246 Available only for systems with CONFIG_HIGHMEM enabled (32b systems). 247 248 This parameter controls whether the high memory is considered for dirty 249 writers throttling. This is not the case by default which means that 250 only the amount of memory directly visible/usable by the kernel can 251 be dirtied. As a result, on systems with a large amount of memory and 252 lowmem basically depleted writers might be throttled too early and 253 streaming writes can get very slow. 254 255 Changing the value to non zero would allow more memory to be dirtied 256 and thus allow writers to write more data which can be flushed to the 257 storage more effectively. Note this also comes with a risk of pre-mature 258 OOM killer because some writers (e.g. direct block device writes) can 259 only use the low memory and they can fill it up with dirty data without 260 any throttling. 261 262 ============================================================== 263 264 hugepages_treat_as_movable 265 266 This parameter controls whether we can allocate hugepages from ZONE_MOVABLE 267 or not. If set to non-zero, hugepages can be allocated from ZONE_MOVABLE. 268 ZONE_MOVABLE is created when kernel boot parameter kernelcore= is specified, 269 so this parameter has no effect if used without kernelcore=. 270 271 Hugepage migration is now available in some situations which depend on the 272 architecture and/or the hugepage size. If a hugepage supports migration, 273 allocation from ZONE_MOVABLE is always enabled for the hugepage regardless 274 of the value of this parameter. 275 IOW, this parameter affects only non-migratable hugepages. 276 277 Assuming that hugepages are not migratable in your system, one usecase of 278 this parameter is that users can make hugepage pool more extensible by 279 enabling the allocation from ZONE_MOVABLE. This is because on ZONE_MOVABLE 280 page reclaim/migration/compaction work more and you can get contiguous 281 memory more likely. Note that using ZONE_MOVABLE for non-migratable 282 hugepages can do harm to other features like memory hotremove (because 283 memory hotremove expects that memory blocks on ZONE_MOVABLE are always 284 removable,) so it's a trade-off responsible for the users. 285 286 ============================================================== 287 288 hugetlb_shm_group 289 290 hugetlb_shm_group contains group id that is allowed to create SysV 291 shared memory segment using hugetlb page. 292 293 ============================================================== 294 295 laptop_mode 296 297 laptop_mode is a knob that controls "laptop mode". All the things that are 298 controlled by this knob are discussed in Documentation/laptops/laptop-mode.txt. 299 300 ============================================================== 301 302 legacy_va_layout 303 304 If non-zero, this sysctl disables the new 32-bit mmap layout - the kernel 305 will use the legacy (2.4) layout for all processes. 306 307 ============================================================== 308 309 lowmem_reserve_ratio 310 311 For some specialised workloads on highmem machines it is dangerous for 312 the kernel to allow process memory to be allocated from the "lowmem" 313 zone. This is because that memory could then be pinned via the mlock() 314 system call, or by unavailability of swapspace. 315 316 And on large highmem machines this lack of reclaimable lowmem memory 317 can be fatal. 318 319 So the Linux page allocator has a mechanism which prevents allocations 320 which _could_ use highmem from using too much lowmem. This means that 321 a certain amount of lowmem is defended from the possibility of being 322 captured into pinned user memory. 323 324 (The same argument applies to the old 16 megabyte ISA DMA region. This 325 mechanism will also defend that region from allocations which could use 326 highmem or lowmem). 327 328 The `lowmem_reserve_ratio' tunable determines how aggressive the kernel is 329 in defending these lower zones. 330 331 If you have a machine which uses highmem or ISA DMA and your 332 applications are using mlock(), or if you are running with no swap then 333 you probably should change the lowmem_reserve_ratio setting. 334 335 The lowmem_reserve_ratio is an array. You can see them by reading this file. 336 - 337 % cat /proc/sys/vm/lowmem_reserve_ratio 338 256 256 32 339 - 340 Note: # of this elements is one fewer than number of zones. Because the highest 341 zone's value is not necessary for following calculation. 342 343 But, these values are not used directly. The kernel calculates # of protection 344 pages for each zones from them. These are shown as array of protection pages 345 in /proc/zoneinfo like followings. (This is an example of x86-64 box). 346 Each zone has an array of protection pages like this. 347 348 - 349 Node 0, zone DMA 350 pages free 1355 351 min 3 352 low 3 353 high 4 354 : 355 : 356 numa_other 0 357 protection: (0, 2004, 2004, 2004) 358 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 359 pagesets 360 cpu: 0 pcp: 0 361 : 362 - 363 These protections are added to score to judge whether this zone should be used 364 for page allocation or should be reclaimed. 365 366 In this example, if normal pages (index=2) are required to this DMA zone and 367 watermark[WMARK_HIGH] is used for watermark, the kernel judges this zone should 368 not be used because pages_free(1355) is smaller than watermark + protection 369 (4 + 2004 = 2008). If this protection value is 0, this zone would be used for 370 normal page requirement. If requirement is DMA zone(index=0), protection 371 (=0) is used. 372 373 zone[i]'s protection[j] is calculated by following expression. 374 375 (i < j): 376 zone[i]->protection[j] 377 = (total sums of managed_pages from zone[i+1] to zone[j] on the node) 378 / lowmem_reserve_ratio[i]; 379 (i = j): 380 (should not be protected. = 0; 381 (i > j): 382 (not necessary, but looks 0) 383 384 The default values of lowmem_reserve_ratio[i] are 385 256 (if zone[i] means DMA or DMA32 zone) 386 32 (others). 387 As above expression, they are reciprocal number of ratio. 388 256 means 1/256. # of protection pages becomes about "0.39%" of total managed 389 pages of higher zones on the node. 390 391 If you would like to protect more pages, smaller values are effective. 392 The minimum value is 1 (1/1 -> 100%). 393 394 ============================================================== 395 396 max_map_count: 397 398 This file contains the maximum number of memory map areas a process 399 may have. Memory map areas are used as a side-effect of calling 400 malloc, directly by mmap, mprotect, and madvise, and also when loading 401 shared libraries. 402 403 While most applications need less than a thousand maps, certain 404 programs, particularly malloc debuggers, may consume lots of them, 405 e.g., up to one or two maps per allocation. 406 407 The default value is 65536. 408 409 ============================================================= 410 411 memory_failure_early_kill: 412 413 Control how to kill processes when uncorrected memory error (typically 414 a 2bit error in a memory module) is detected in the background by hardware 415 that cannot be handled by the kernel. In some cases (like the page 416 still having a valid copy on disk) the kernel will handle the failure 417 transparently without affecting any applications. But if there is 418 no other uptodate copy of the data it will kill to prevent any data 419 corruptions from propagating. 420 421 1: Kill all processes that have the corrupted and not reloadable page mapped 422 as soon as the corruption is detected. Note this is not supported 423 for a few types of pages, like kernel internally allocated data or 424 the swap cache, but works for the majority of user pages. 425 426 0: Only unmap the corrupted page from all processes and only kill a process 427 who tries to access it. 428 429 The kill is done using a catchable SIGBUS with BUS_MCEERR_AO, so processes can 430 handle this if they want to. 431 432 This is only active on architectures/platforms with advanced machine 433 check handling and depends on the hardware capabilities. 434 435 Applications can override this setting individually with the PR_MCE_KILL prctl 436 437 ============================================================== 438 439 memory_failure_recovery 440 441 Enable memory failure recovery (when supported by the platform) 442 443 1: Attempt recovery. 444 445 0: Always panic on a memory failure. 446 447 ============================================================== 448 449 min_free_kbytes: 450 451 This is used to force the Linux VM to keep a minimum number 452 of kilobytes free. The VM uses this number to compute a 453 watermark[WMARK_MIN] value for each lowmem zone in the system. 454 Each lowmem zone gets a number of reserved free pages based 455 proportionally on its size. 456 457 Some minimal amount of memory is needed to satisfy PF_MEMALLOC 458 allocations; if you set this to lower than 1024KB, your system will 459 become subtly broken, and prone to deadlock under high loads. 460 461 Setting this too high will OOM your machine instantly. 462 463 ============================================================= 464 465 min_slab_ratio: 466 467 This is available only on NUMA kernels. 468 469 A percentage of the total pages in each zone. On Zone reclaim 470 (fallback from the local zone occurs) slabs will be reclaimed if more 471 than this percentage of pages in a zone are reclaimable slab pages. 472 This insures that the slab growth stays under control even in NUMA 473 systems that rarely perform global reclaim. 474 475 The default is 5 percent. 476 477 Note that slab reclaim is triggered in a per zone / node fashion. 478 The process of reclaiming slab memory is currently not node specific 479 and may not be fast. 480 481 ============================================================= 482 483 min_unmapped_ratio: 484 485 This is available only on NUMA kernels. 486 487 This is a percentage of the total pages in each zone. Zone reclaim will 488 only occur if more than this percentage of pages are in a state that 489 zone_reclaim_mode allows to be reclaimed. 490 491 If zone_reclaim_mode has the value 4 OR'd, then the percentage is compared 492 against all file-backed unmapped pages including swapcache pages and tmpfs 493 files. Otherwise, only unmapped pages backed by normal files but not tmpfs 494 files and similar are considered. 495 496 The default is 1 percent. 497 498 ============================================================== 499 500 mmap_min_addr 501 502 This file indicates the amount of address space which a user process will 503 be restricted from mmapping. Since kernel null dereference bugs could 504 accidentally operate based on the information in the first couple of pages 505 of memory userspace processes should not be allowed to write to them. By 506 default this value is set to 0 and no protections will be enforced by the 507 security module. Setting this value to something like 64k will allow the 508 vast majority of applications to work correctly and provide defense in depth 509 against future potential kernel bugs. 510 511 ============================================================== 512 513 mmap_rnd_bits: 514 515 This value can be used to select the number of bits to use to 516 determine the random offset to the base address of vma regions 517 resulting from mmap allocations on architectures which support 518 tuning address space randomization. This value will be bounded 519 by the architecture's minimum and maximum supported values. 520 521 This value can be changed after boot using the 522 /proc/sys/vm/mmap_rnd_bits tunable 523 524 ============================================================== 525 526 mmap_rnd_compat_bits: 527 528 This value can be used to select the number of bits to use to 529 determine the random offset to the base address of vma regions 530 resulting from mmap allocations for applications run in 531 compatibility mode on architectures which support tuning address 532 space randomization. This value will be bounded by the 533 architecture's minimum and maximum supported values. 534 535 This value can be changed after boot using the 536 /proc/sys/vm/mmap_rnd_compat_bits tunable 537 538 ============================================================== 539 540 nr_hugepages 541 542 Change the minimum size of the hugepage pool. 543 544 See Documentation/vm/hugetlbpage.txt 545 546 ============================================================== 547 548 nr_overcommit_hugepages 549 550 Change the maximum size of the hugepage pool. The maximum is 551 nr_hugepages + nr_overcommit_hugepages. 552 553 See Documentation/vm/hugetlbpage.txt 554 555 ============================================================== 556 557 nr_trim_pages 558 559 This is available only on NOMMU kernels. 560 561 This value adjusts the excess page trimming behaviour of power-of-2 aligned 562 NOMMU mmap allocations. 563 564 A value of 0 disables trimming of allocations entirely, while a value of 1 565 trims excess pages aggressively. Any value >= 1 acts as the watermark where 566 trimming of allocations is initiated. 567 568 The default value is 1. 569 570 See Documentation/nommu-mmap.txt for more information. 571 572 ============================================================== 573 574 numa_zonelist_order 575 576 This sysctl is only for NUMA and it is deprecated. Anything but 577 Node order will fail! 578 579 'where the memory is allocated from' is controlled by zonelists. 580 (This documentation ignores ZONE_HIGHMEM/ZONE_DMA32 for simple explanation. 581 you may be able to read ZONE_DMA as ZONE_DMA32...) 582 583 In non-NUMA case, a zonelist for GFP_KERNEL is ordered as following. 584 ZONE_NORMAL -> ZONE_DMA 585 This means that a memory allocation request for GFP_KERNEL will 586 get memory from ZONE_DMA only when ZONE_NORMAL is not available. 587 588 In NUMA case, you can think of following 2 types of order. 589 Assume 2 node NUMA and below is zonelist of Node(0)'s GFP_KERNEL 590 591 (A) Node(0) ZONE_NORMAL -> Node(0) ZONE_DMA -> Node(1) ZONE_NORMAL 592 (B) Node(0) ZONE_NORMAL -> Node(1) ZONE_NORMAL -> Node(0) ZONE_DMA. 593 594 Type(A) offers the best locality for processes on Node(0), but ZONE_DMA 595 will be used before ZONE_NORMAL exhaustion. This increases possibility of 596 out-of-memory(OOM) of ZONE_DMA because ZONE_DMA is tend to be small. 597 598 Type(B) cannot offer the best locality but is more robust against OOM of 599 the DMA zone. 600 601 Type(A) is called as "Node" order. Type (B) is "Zone" order. 602 603 "Node order" orders the zonelists by node, then by zone within each node. 604 Specify "[Nn]ode" for node order 605 606 "Zone Order" orders the zonelists by zone type, then by node within each 607 zone. Specify "[Zz]one" for zone order. 608 609 Specify "[Dd]efault" to request automatic configuration. 610 611 On 32-bit, the Normal zone needs to be preserved for allocations accessible 612 by the kernel, so "zone" order will be selected. 613 614 On 64-bit, devices that require DMA32/DMA are relatively rare, so "node" 615 order will be selected. 616 617 Default order is recommended unless this is causing problems for your 618 system/application. 619 620 ============================================================== 621 622 oom_dump_tasks 623 624 Enables a system-wide task dump (excluding kernel threads) to be produced 625 when the kernel performs an OOM-killing and includes such information as 626 pid, uid, tgid, vm size, rss, pgtables_bytes, swapents, oom_score_adj 627 score, and name. This is helpful to determine why the OOM killer was 628 invoked, to identify the rogue task that caused it, and to determine why 629 the OOM killer chose the task it did to kill. 630 631 If this is set to zero, this information is suppressed. On very 632 large systems with thousands of tasks it may not be feasible to dump 633 the memory state information for each one. Such systems should not 634 be forced to incur a performance penalty in OOM conditions when the 635 information may not be desired. 636 637 If this is set to non-zero, this information is shown whenever the 638 OOM killer actually kills a memory-hogging task. 639 640 The default value is 1 (enabled). 641 642 ============================================================== 643 644 oom_kill_allocating_task 645 646 This enables or disables killing the OOM-triggering task in 647 out-of-memory situations. 648 649 If this is set to zero, the OOM killer will scan through the entire 650 tasklist and select a task based on heuristics to kill. This normally 651 selects a rogue memory-hogging task that frees up a large amount of 652 memory when killed. 653 654 If this is set to non-zero, the OOM killer simply kills the task that 655 triggered the out-of-memory condition. This avoids the expensive 656 tasklist scan. 657 658 If panic_on_oom is selected, it takes precedence over whatever value 659 is used in oom_kill_allocating_task. 660 661 The default value is 0. 662 663 ============================================================== 664 665 overcommit_kbytes: 666 667 When overcommit_memory is set to 2, the committed address space is not 668 permitted to exceed swap plus this amount of physical RAM. See below. 669 670 Note: overcommit_kbytes is the counterpart of overcommit_ratio. Only one 671 of them may be specified at a time. Setting one disables the other (which 672 then appears as 0 when read). 673 674 ============================================================== 675 676 overcommit_memory: 677 678 This value contains a flag that enables memory overcommitment. 679 680 When this flag is 0, the kernel attempts to estimate the amount 681 of free memory left when userspace requests more memory. 682 683 When this flag is 1, the kernel pretends there is always enough 684 memory until it actually runs out. 685 686 When this flag is 2, the kernel uses a "never overcommit" 687 policy that attempts to prevent any overcommit of memory. 688 Note that user_reserve_kbytes affects this policy. 689 690 This feature can be very useful because there are a lot of 691 programs that malloc() huge amounts of memory "just-in-case" 692 and don't use much of it. 693 694 The default value is 0. 695 696 See Documentation/vm/overcommit-accounting and 697 mm/mmap.c::__vm_enough_memory() for more information. 698 699 ============================================================== 700 701 overcommit_ratio: 702 703 When overcommit_memory is set to 2, the committed address 704 space is not permitted to exceed swap plus this percentage 705 of physical RAM. See above. 706 707 ============================================================== 708 709 page-cluster 710 711 page-cluster controls the number of pages up to which consecutive pages 712 are read in from swap in a single attempt. This is the swap counterpart 713 to page cache readahead. 714 The mentioned consecutivity is not in terms of virtual/physical addresses, 715 but consecutive on swap space - that means they were swapped out together. 716 717 It is a logarithmic value - setting it to zero means "1 page", setting 718 it to 1 means "2 pages", setting it to 2 means "4 pages", etc. 719 Zero disables swap readahead completely. 720 721 The default value is three (eight pages at a time). There may be some 722 small benefits in tuning this to a different value if your workload is 723 swap-intensive. 724 725 Lower values mean lower latencies for initial faults, but at the same time 726 extra faults and I/O delays for following faults if they would have been part of 727 that consecutive pages readahead would have brought in. 728 729 ============================================================= 730 731 panic_on_oom 732 733 This enables or disables panic on out-of-memory feature. 734 735 If this is set to 0, the kernel will kill some rogue process, 736 called oom_killer. Usually, oom_killer can kill rogue processes and 737 system will survive. 738 739 If this is set to 1, the kernel panics when out-of-memory happens. 740 However, if a process limits using nodes by mempolicy/cpusets, 741 and those nodes become memory exhaustion status, one process 742 may be killed by oom-killer. No panic occurs in this case. 743 Because other nodes' memory may be free. This means system total status 744 may be not fatal yet. 745 746 If this is set to 2, the kernel panics compulsorily even on the 747 above-mentioned. Even oom happens under memory cgroup, the whole 748 system panics. 749 750 The default value is 0. 751 1 and 2 are for failover of clustering. Please select either 752 according to your policy of failover. 753 panic_on_oom=2+kdump gives you very strong tool to investigate 754 why oom happens. You can get snapshot. 755 756 ============================================================= 757 758 percpu_pagelist_fraction 759 760 This is the fraction of pages at most (high mark pcp->high) in each zone that 761 are allocated for each per cpu page list. The min value for this is 8. It 762 means that we don't allow more than 1/8th of pages in each zone to be 763 allocated in any single per_cpu_pagelist. This entry only changes the value 764 of hot per cpu pagelists. User can specify a number like 100 to allocate 765 1/100th of each zone to each per cpu page list. 766 767 The batch value of each per cpu pagelist is also updated as a result. It is 768 set to pcp->high/4. The upper limit of batch is (PAGE_SHIFT * 8) 769 770 The initial value is zero. Kernel does not use this value at boot time to set 771 the high water marks for each per cpu page list. If the user writes '0' to this 772 sysctl, it will revert to this default behavior. 773 774 ============================================================== 775 776 stat_interval 777 778 The time interval between which vm statistics are updated. The default 779 is 1 second. 780 781 ============================================================== 782 783 stat_refresh 784 785 Any read or write (by root only) flushes all the per-cpu vm statistics 786 into their global totals, for more accurate reports when testing 787 e.g. cat /proc/sys/vm/stat_refresh /proc/meminfo 788 789 As a side-effect, it also checks for negative totals (elsewhere reported 790 as 0) and "fails" with EINVAL if any are found, with a warning in dmesg. 791 (At time of writing, a few stats are known sometimes to be found negative, 792 with no ill effects: errors and warnings on these stats are suppressed.) 793 794 ============================================================== 795 796 numa_stat 797 798 This interface allows runtime configuration of numa statistics. 799 800 When page allocation performance becomes a bottleneck and you can tolerate 801 some possible tool breakage and decreased numa counter precision, you can 802 do: 803 echo 0 > /proc/sys/vm/numa_stat 804 805 When page allocation performance is not a bottleneck and you want all 806 tooling to work, you can do: 807 echo 1 > /proc/sys/vm/numa_stat 808 809 ============================================================== 810 811 swappiness 812 813 This control is used to define how aggressive the kernel will swap 814 memory pages. Higher values will increase aggressiveness, lower values 815 decrease the amount of swap. A value of 0 instructs the kernel not to 816 initiate swap until the amount of free and file-backed pages is less 817 than the high water mark in a zone. 818 819 The default value is 60. 820 821 ============================================================== 822 823 - user_reserve_kbytes 824 825 When overcommit_memory is set to 2, "never overcommit" mode, reserve 826 min(3% of current process size, user_reserve_kbytes) of free memory. 827 This is intended to prevent a user from starting a single memory hogging 828 process, such that they cannot recover (kill the hog). 829 830 user_reserve_kbytes defaults to min(3% of the current process size, 128MB). 831 832 If this is reduced to zero, then the user will be allowed to allocate 833 all free memory with a single process, minus admin_reserve_kbytes. 834 Any subsequent attempts to execute a command will result in 835 "fork: Cannot allocate memory". 836 837 Changing this takes effect whenever an application requests memory. 838 839 ============================================================== 840 841 vfs_cache_pressure 842 ------------------ 843 844 This percentage value controls the tendency of the kernel to reclaim 845 the memory which is used for caching of directory and inode objects. 846 847 At the default value of vfs_cache_pressure=100 the kernel will attempt to 848 reclaim dentries and inodes at a "fair" rate with respect to pagecache and 849 swapcache reclaim. Decreasing vfs_cache_pressure causes the kernel to prefer 850 to retain dentry and inode caches. When vfs_cache_pressure=0, the kernel will 851 never reclaim dentries and inodes due to memory pressure and this can easily 852 lead to out-of-memory conditions. Increasing vfs_cache_pressure beyond 100 853 causes the kernel to prefer to reclaim dentries and inodes. 854 855 Increasing vfs_cache_pressure significantly beyond 100 may have negative 856 performance impact. Reclaim code needs to take various locks to find freeable 857 directory and inode objects. With vfs_cache_pressure=1000, it will look for 858 ten times more freeable objects than there are. 859 860 ============================================================= 861 862 watermark_scale_factor: 863 864 This factor controls the aggressiveness of kswapd. It defines the 865 amount of memory left in a node/system before kswapd is woken up and 866 how much memory needs to be free before kswapd goes back to sleep. 867 868 The unit is in fractions of 10,000. The default value of 10 means the 869 distances between watermarks are 0.1% of the available memory in the 870 node/system. The maximum value is 1000, or 10% of memory. 871 872 A high rate of threads entering direct reclaim (allocstall) or kswapd 873 going to sleep prematurely (kswapd_low_wmark_hit_quickly) can indicate 874 that the number of free pages kswapd maintains for latency reasons is 875 too small for the allocation bursts occurring in the system. This knob 876 can then be used to tune kswapd aggressiveness accordingly. 877 878 ============================================================== 879 880 zone_reclaim_mode: 881 882 Zone_reclaim_mode allows someone to set more or less aggressive approaches to 883 reclaim memory when a zone runs out of memory. If it is set to zero then no 884 zone reclaim occurs. Allocations will be satisfied from other zones / nodes 885 in the system. 886 887 This is value ORed together of 888 889 1 = Zone reclaim on 890 2 = Zone reclaim writes dirty pages out 891 4 = Zone reclaim swaps pages 892 893 zone_reclaim_mode is disabled by default. For file servers or workloads 894 that benefit from having their data cached, zone_reclaim_mode should be 895 left disabled as the caching effect is likely to be more important than 896 data locality. 897 898 zone_reclaim may be enabled if it's known that the workload is partitioned 899 such that each partition fits within a NUMA node and that accessing remote 900 memory would cause a measurable performance reduction. The page allocator 901 will then reclaim easily reusable pages (those page cache pages that are 902 currently not used) before allocating off node pages. 903 904 Allowing zone reclaim to write out pages stops processes that are 905 writing large amounts of data from dirtying pages on other nodes. Zone 906 reclaim will write out dirty pages if a zone fills up and so effectively 907 throttle the process. This may decrease the performance of a single process 908 since it cannot use all of system memory to buffer the outgoing writes 909 anymore but it preserve the memory on other nodes so that the performance 910 of other processes running on other nodes will not be affected. 911 912 Allowing regular swap effectively restricts allocations to the local 913 node unless explicitly overridden by memory policies or cpuset 914 configurations. 915 916 ============ End of Document =================================