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- nr_hugepages 46- nr_overcommit_hugepages 47- nr_trim_pages (only if CONFIG_MMU=n) 48- numa_zonelist_order 49- oom_dump_tasks 50- oom_kill_allocating_task 51- overcommit_kbytes 52- overcommit_memory 53- overcommit_ratio 54- page-cluster 55- panic_on_oom 56- percpu_pagelist_fraction 57- stat_interval 58- swappiness 59- user_reserve_kbytes 60- vfs_cache_pressure 61- zone_reclaim_mode 62 63============================================================== 64 65admin_reserve_kbytes 66 67The amount of free memory in the system that should be reserved for users 68with the capability cap_sys_admin. 69 70admin_reserve_kbytes defaults to min(3% of free pages, 8MB) 71 72That should provide enough for the admin to log in and kill a process, 73if necessary, under the default overcommit 'guess' mode. 74 75Systems running under overcommit 'never' should increase this to account 76for the full Virtual Memory Size of programs used to recover. Otherwise, 77root may not be able to log in to recover the system. 78 79How do you calculate a minimum useful reserve? 80 81sshd or login + bash (or some other shell) + top (or ps, kill, etc.) 82 83For overcommit 'guess', we can sum resident set sizes (RSS). 84On x86_64 this is about 8MB. 85 86For overcommit 'never', we can take the max of their virtual sizes (VSZ) 87and add the sum of their RSS. 88On x86_64 this is about 128MB. 89 90Changing this takes effect whenever an application requests memory. 91 92============================================================== 93 94block_dump 95 96block_dump enables block I/O debugging when set to a nonzero value. More 97information on block I/O debugging is in Documentation/laptops/laptop-mode.txt. 98 99============================================================== 100 101compact_memory 102 103Available only when CONFIG_COMPACTION is set. When 1 is written to the file, 104all zones are compacted such that free memory is available in contiguous 105blocks where possible. This can be important for example in the allocation of 106huge pages although processes will also directly compact memory as required. 107 108============================================================== 109 110compact_unevictable_allowed 111 112Available only when CONFIG_COMPACTION is set. When set to 1, compaction is 113allowed to examine the unevictable lru (mlocked pages) for pages to compact. 114This should be used on systems where stalls for minor page faults are an 115acceptable trade for large contiguous free memory. Set to 0 to prevent 116compaction from moving pages that are unevictable. Default value is 1. 117 118============================================================== 119 120dirty_background_bytes 121 122Contains the amount of dirty memory at which the background kernel 123flusher threads will start writeback. 124 125Note: dirty_background_bytes is the counterpart of dirty_background_ratio. Only 126one of them may be specified at a time. When one sysctl is written it is 127immediately taken into account to evaluate the dirty memory limits and the 128other appears as 0 when read. 129 130============================================================== 131 132dirty_background_ratio 133 134Contains, as a percentage of total available memory that contains free pages 135and reclaimable pages, the number of pages at which the background kernel 136flusher threads will start writing out dirty data. 137 138The total avaiable memory is not equal to total system memory. 139 140============================================================== 141 142dirty_bytes 143 144Contains the amount of dirty memory at which a process generating disk writes 145will itself start writeback. 146 147Note: dirty_bytes is the counterpart of dirty_ratio. Only one of them may be 148specified at a time. When one sysctl is written it is immediately taken into 149account to evaluate the dirty memory limits and the other appears as 0 when 150read. 151 152Note: the minimum value allowed for dirty_bytes is two pages (in bytes); any 153value lower than this limit will be ignored and the old configuration will be 154retained. 155 156============================================================== 157 158dirty_expire_centisecs 159 160This tunable is used to define when dirty data is old enough to be eligible 161for writeout by the kernel flusher threads. It is expressed in 100'ths 162of a second. Data which has been dirty in-memory for longer than this 163interval will be written out next time a flusher thread wakes up. 164 165============================================================== 166 167dirty_ratio 168 169Contains, as a percentage of total available memory that contains free pages 170and reclaimable pages, the number of pages at which a process which is 171generating disk writes will itself start writing out dirty data. 172 173The total avaiable memory is not equal to total system memory. 174 175============================================================== 176 177dirty_writeback_centisecs 178 179The kernel flusher threads will periodically wake up and write `old' data 180out to disk. This tunable expresses the interval between those wakeups, in 181100'ths of a second. 182 183Setting this to zero disables periodic writeback altogether. 184 185============================================================== 186 187drop_caches 188 189Writing to this will cause the kernel to drop clean caches, as well as 190reclaimable slab objects like dentries and inodes. Once dropped, their 191memory becomes free. 192 193To free pagecache: 194 echo 1 > /proc/sys/vm/drop_caches 195To free reclaimable slab objects (includes dentries and inodes): 196 echo 2 > /proc/sys/vm/drop_caches 197To free slab objects and pagecache: 198 echo 3 > /proc/sys/vm/drop_caches 199 200This is a non-destructive operation and will not free any dirty objects. 201To increase the number of objects freed by this operation, the user may run 202`sync' prior to writing to /proc/sys/vm/drop_caches. This will minimize the 203number of dirty objects on the system and create more candidates to be 204dropped. 205 206This file is not a means to control the growth of the various kernel caches 207(inodes, dentries, pagecache, etc...) These objects are automatically 208reclaimed by the kernel when memory is needed elsewhere on the system. 209 210Use of this file can cause performance problems. Since it discards cached 211objects, it may cost a significant amount of I/O and CPU to recreate the 212dropped objects, especially if they were under heavy use. Because of this, 213use outside of a testing or debugging environment is not recommended. 214 215You may see informational messages in your kernel log when this file is 216used: 217 218 cat (1234): drop_caches: 3 219 220These are informational only. They do not mean that anything is wrong 221with your system. To disable them, echo 4 (bit 3) into drop_caches. 222 223============================================================== 224 225extfrag_threshold 226 227This parameter affects whether the kernel will compact memory or direct 228reclaim to satisfy a high-order allocation. The extfrag/extfrag_index file in 229debugfs shows what the fragmentation index for each order is in each zone in 230the system. Values tending towards 0 imply allocations would fail due to lack 231of memory, values towards 1000 imply failures are due to fragmentation and -1 232implies that the allocation will succeed as long as watermarks are met. 233 234The kernel will not compact memory in a zone if the 235fragmentation index is <= extfrag_threshold. The default value is 500. 236 237============================================================== 238 239hugepages_treat_as_movable 240 241This parameter controls whether we can allocate hugepages from ZONE_MOVABLE 242or not. If set to non-zero, hugepages can be allocated from ZONE_MOVABLE. 243ZONE_MOVABLE is created when kernel boot parameter kernelcore= is specified, 244so this parameter has no effect if used without kernelcore=. 245 246Hugepage migration is now available in some situations which depend on the 247architecture and/or the hugepage size. If a hugepage supports migration, 248allocation from ZONE_MOVABLE is always enabled for the hugepage regardless 249of the value of this parameter. 250IOW, this parameter affects only non-migratable hugepages. 251 252Assuming that hugepages are not migratable in your system, one usecase of 253this parameter is that users can make hugepage pool more extensible by 254enabling the allocation from ZONE_MOVABLE. This is because on ZONE_MOVABLE 255page reclaim/migration/compaction work more and you can get contiguous 256memory more likely. Note that using ZONE_MOVABLE for non-migratable 257hugepages can do harm to other features like memory hotremove (because 258memory hotremove expects that memory blocks on ZONE_MOVABLE are always 259removable,) so it's a trade-off responsible for the users. 260 261============================================================== 262 263hugetlb_shm_group 264 265hugetlb_shm_group contains group id that is allowed to create SysV 266shared memory segment using hugetlb page. 267 268============================================================== 269 270laptop_mode 271 272laptop_mode is a knob that controls "laptop mode". All the things that are 273controlled by this knob are discussed in Documentation/laptops/laptop-mode.txt. 274 275============================================================== 276 277legacy_va_layout 278 279If non-zero, this sysctl disables the new 32-bit mmap layout - the kernel 280will use the legacy (2.4) layout for all processes. 281 282============================================================== 283 284lowmem_reserve_ratio 285 286For some specialised workloads on highmem machines it is dangerous for 287the kernel to allow process memory to be allocated from the "lowmem" 288zone. This is because that memory could then be pinned via the mlock() 289system call, or by unavailability of swapspace. 290 291And on large highmem machines this lack of reclaimable lowmem memory 292can be fatal. 293 294So the Linux page allocator has a mechanism which prevents allocations 295which _could_ use highmem from using too much lowmem. This means that 296a certain amount of lowmem is defended from the possibility of being 297captured into pinned user memory. 298 299(The same argument applies to the old 16 megabyte ISA DMA region. This 300mechanism will also defend that region from allocations which could use 301highmem or lowmem). 302 303The `lowmem_reserve_ratio' tunable determines how aggressive the kernel is 304in defending these lower zones. 305 306If you have a machine which uses highmem or ISA DMA and your 307applications are using mlock(), or if you are running with no swap then 308you probably should change the lowmem_reserve_ratio setting. 309 310The lowmem_reserve_ratio is an array. You can see them by reading this file. 311- 312% cat /proc/sys/vm/lowmem_reserve_ratio 313256 256 32 314- 315Note: # of this elements is one fewer than number of zones. Because the highest 316 zone's value is not necessary for following calculation. 317 318But, these values are not used directly. The kernel calculates # of protection 319pages for each zones from them. These are shown as array of protection pages 320in /proc/zoneinfo like followings. (This is an example of x86-64 box). 321Each zone has an array of protection pages like this. 322 323- 324Node 0, zone DMA 325 pages free 1355 326 min 3 327 low 3 328 high 4 329 : 330 : 331 numa_other 0 332 protection: (0, 2004, 2004, 2004) 333 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 334 pagesets 335 cpu: 0 pcp: 0 336 : 337- 338These protections are added to score to judge whether this zone should be used 339for page allocation or should be reclaimed. 340 341In this example, if normal pages (index=2) are required to this DMA zone and 342watermark[WMARK_HIGH] is used for watermark, the kernel judges this zone should 343not be used because pages_free(1355) is smaller than watermark + protection[2] 344(4 + 2004 = 2008). If this protection value is 0, this zone would be used for 345normal page requirement. If requirement is DMA zone(index=0), protection[0] 346(=0) is used. 347 348zone[i]'s protection[j] is calculated by following expression. 349 350(i < j): 351 zone[i]->protection[j] 352 = (total sums of managed_pages from zone[i+1] to zone[j] on the node) 353 / lowmem_reserve_ratio[i]; 354(i = j): 355 (should not be protected. = 0; 356(i > j): 357 (not necessary, but looks 0) 358 359The default values of lowmem_reserve_ratio[i] are 360 256 (if zone[i] means DMA or DMA32 zone) 361 32 (others). 362As above expression, they are reciprocal number of ratio. 363256 means 1/256. # of protection pages becomes about "0.39%" of total managed 364pages of higher zones on the node. 365 366If you would like to protect more pages, smaller values are effective. 367The minimum value is 1 (1/1 -> 100%). 368 369============================================================== 370 371max_map_count: 372 373This file contains the maximum number of memory map areas a process 374may have. Memory map areas are used as a side-effect of calling 375malloc, directly by mmap and mprotect, and also when loading shared 376libraries. 377 378While most applications need less than a thousand maps, certain 379programs, particularly malloc debuggers, may consume lots of them, 380e.g., up to one or two maps per allocation. 381 382The default value is 65536. 383 384============================================================= 385 386memory_failure_early_kill: 387 388Control how to kill processes when uncorrected memory error (typically 389a 2bit error in a memory module) is detected in the background by hardware 390that cannot be handled by the kernel. In some cases (like the page 391still having a valid copy on disk) the kernel will handle the failure 392transparently without affecting any applications. But if there is 393no other uptodate copy of the data it will kill to prevent any data 394corruptions from propagating. 395 3961: Kill all processes that have the corrupted and not reloadable page mapped 397as soon as the corruption is detected. Note this is not supported 398for a few types of pages, like kernel internally allocated data or 399the swap cache, but works for the majority of user pages. 400 4010: Only unmap the corrupted page from all processes and only kill a process 402who tries to access it. 403 404The kill is done using a catchable SIGBUS with BUS_MCEERR_AO, so processes can 405handle this if they want to. 406 407This is only active on architectures/platforms with advanced machine 408check handling and depends on the hardware capabilities. 409 410Applications can override this setting individually with the PR_MCE_KILL prctl 411 412============================================================== 413 414memory_failure_recovery 415 416Enable memory failure recovery (when supported by the platform) 417 4181: Attempt recovery. 419 4200: Always panic on a memory failure. 421 422============================================================== 423 424min_free_kbytes: 425 426This is used to force the Linux VM to keep a minimum number 427of kilobytes free. The VM uses this number to compute a 428watermark[WMARK_MIN] value for each lowmem zone in the system. 429Each lowmem zone gets a number of reserved free pages based 430proportionally on its size. 431 432Some minimal amount of memory is needed to satisfy PF_MEMALLOC 433allocations; if you set this to lower than 1024KB, your system will 434become subtly broken, and prone to deadlock under high loads. 435 436Setting this too high will OOM your machine instantly. 437 438============================================================= 439 440min_slab_ratio: 441 442This is available only on NUMA kernels. 443 444A percentage of the total pages in each zone. On Zone reclaim 445(fallback from the local zone occurs) slabs will be reclaimed if more 446than this percentage of pages in a zone are reclaimable slab pages. 447This insures that the slab growth stays under control even in NUMA 448systems that rarely perform global reclaim. 449 450The default is 5 percent. 451 452Note that slab reclaim is triggered in a per zone / node fashion. 453The process of reclaiming slab memory is currently not node specific 454and may not be fast. 455 456============================================================= 457 458min_unmapped_ratio: 459 460This is available only on NUMA kernels. 461 462This is a percentage of the total pages in each zone. Zone reclaim will 463only occur if more than this percentage of pages are in a state that 464zone_reclaim_mode allows to be reclaimed. 465 466If zone_reclaim_mode has the value 4 OR'd, then the percentage is compared 467against all file-backed unmapped pages including swapcache pages and tmpfs 468files. Otherwise, only unmapped pages backed by normal files but not tmpfs 469files and similar are considered. 470 471The default is 1 percent. 472 473============================================================== 474 475mmap_min_addr 476 477This file indicates the amount of address space which a user process will 478be restricted from mmapping. Since kernel null dereference bugs could 479accidentally operate based on the information in the first couple of pages 480of memory userspace processes should not be allowed to write to them. By 481default this value is set to 0 and no protections will be enforced by the 482security module. Setting this value to something like 64k will allow the 483vast majority of applications to work correctly and provide defense in depth 484against future potential kernel bugs. 485 486============================================================== 487 488nr_hugepages 489 490Change the minimum size of the hugepage pool. 491 492See Documentation/vm/hugetlbpage.txt 493 494============================================================== 495 496nr_overcommit_hugepages 497 498Change the maximum size of the hugepage pool. The maximum is 499nr_hugepages + nr_overcommit_hugepages. 500 501See Documentation/vm/hugetlbpage.txt 502 503============================================================== 504 505nr_trim_pages 506 507This is available only on NOMMU kernels. 508 509This value adjusts the excess page trimming behaviour of power-of-2 aligned 510NOMMU mmap allocations. 511 512A value of 0 disables trimming of allocations entirely, while a value of 1 513trims excess pages aggressively. Any value >= 1 acts as the watermark where 514trimming of allocations is initiated. 515 516The default value is 1. 517 518See Documentation/nommu-mmap.txt for more information. 519 520============================================================== 521 522numa_zonelist_order 523 524This sysctl is only for NUMA. 525'where the memory is allocated from' is controlled by zonelists. 526(This documentation ignores ZONE_HIGHMEM/ZONE_DMA32 for simple explanation. 527 you may be able to read ZONE_DMA as ZONE_DMA32...) 528 529In non-NUMA case, a zonelist for GFP_KERNEL is ordered as following. 530ZONE_NORMAL -> ZONE_DMA 531This means that a memory allocation request for GFP_KERNEL will 532get memory from ZONE_DMA only when ZONE_NORMAL is not available. 533 534In NUMA case, you can think of following 2 types of order. 535Assume 2 node NUMA and below is zonelist of Node(0)'s GFP_KERNEL 536 537(A) Node(0) ZONE_NORMAL -> Node(0) ZONE_DMA -> Node(1) ZONE_NORMAL 538(B) Node(0) ZONE_NORMAL -> Node(1) ZONE_NORMAL -> Node(0) ZONE_DMA. 539 540Type(A) offers the best locality for processes on Node(0), but ZONE_DMA 541will be used before ZONE_NORMAL exhaustion. This increases possibility of 542out-of-memory(OOM) of ZONE_DMA because ZONE_DMA is tend to be small. 543 544Type(B) cannot offer the best locality but is more robust against OOM of 545the DMA zone. 546 547Type(A) is called as "Node" order. Type (B) is "Zone" order. 548 549"Node order" orders the zonelists by node, then by zone within each node. 550Specify "[Nn]ode" for node order 551 552"Zone Order" orders the zonelists by zone type, then by node within each 553zone. Specify "[Zz]one" for zone order. 554 555Specify "[Dd]efault" to request automatic configuration. Autoconfiguration 556will select "node" order in following case. 557(1) if the DMA zone does not exist or 558(2) if the DMA zone comprises greater than 50% of the available memory or 559(3) if any node's DMA zone comprises greater than 70% of its local memory and 560 the amount of local memory is big enough. 561 562Otherwise, "zone" order will be selected. Default order is recommended unless 563this is causing problems for your system/application. 564 565============================================================== 566 567oom_dump_tasks 568 569Enables a system-wide task dump (excluding kernel threads) to be produced 570when the kernel performs an OOM-killing and includes such information as 571pid, uid, tgid, vm size, rss, nr_ptes, nr_pmds, swapents, oom_score_adj 572score, and name. This is helpful to determine why the OOM killer was 573invoked, to identify the rogue task that caused it, and to determine why 574the OOM killer chose the task it did to kill. 575 576If this is set to zero, this information is suppressed. On very 577large systems with thousands of tasks it may not be feasible to dump 578the memory state information for each one. Such systems should not 579be forced to incur a performance penalty in OOM conditions when the 580information may not be desired. 581 582If this is set to non-zero, this information is shown whenever the 583OOM killer actually kills a memory-hogging task. 584 585The default value is 1 (enabled). 586 587============================================================== 588 589oom_kill_allocating_task 590 591This enables or disables killing the OOM-triggering task in 592out-of-memory situations. 593 594If this is set to zero, the OOM killer will scan through the entire 595tasklist and select a task based on heuristics to kill. This normally 596selects a rogue memory-hogging task that frees up a large amount of 597memory when killed. 598 599If this is set to non-zero, the OOM killer simply kills the task that 600triggered the out-of-memory condition. This avoids the expensive 601tasklist scan. 602 603If panic_on_oom is selected, it takes precedence over whatever value 604is used in oom_kill_allocating_task. 605 606The default value is 0. 607 608============================================================== 609 610overcommit_kbytes: 611 612When overcommit_memory is set to 2, the committed address space is not 613permitted to exceed swap plus this amount of physical RAM. See below. 614 615Note: overcommit_kbytes is the counterpart of overcommit_ratio. Only one 616of them may be specified at a time. Setting one disables the other (which 617then appears as 0 when read). 618 619============================================================== 620 621overcommit_memory: 622 623This value contains a flag that enables memory overcommitment. 624 625When this flag is 0, the kernel attempts to estimate the amount 626of free memory left when userspace requests more memory. 627 628When this flag is 1, the kernel pretends there is always enough 629memory until it actually runs out. 630 631When this flag is 2, the kernel uses a "never overcommit" 632policy that attempts to prevent any overcommit of memory. 633Note that user_reserve_kbytes affects this policy. 634 635This feature can be very useful because there are a lot of 636programs that malloc() huge amounts of memory "just-in-case" 637and don't use much of it. 638 639The default value is 0. 640 641See Documentation/vm/overcommit-accounting and 642mm/mmap.c::__vm_enough_memory() for more information. 643 644============================================================== 645 646overcommit_ratio: 647 648When overcommit_memory is set to 2, the committed address 649space is not permitted to exceed swap plus this percentage 650of physical RAM. See above. 651 652============================================================== 653 654page-cluster 655 656page-cluster controls the number of pages up to which consecutive pages 657are read in from swap in a single attempt. This is the swap counterpart 658to page cache readahead. 659The mentioned consecutivity is not in terms of virtual/physical addresses, 660but consecutive on swap space - that means they were swapped out together. 661 662It is a logarithmic value - setting it to zero means "1 page", setting 663it to 1 means "2 pages", setting it to 2 means "4 pages", etc. 664Zero disables swap readahead completely. 665 666The default value is three (eight pages at a time). There may be some 667small benefits in tuning this to a different value if your workload is 668swap-intensive. 669 670Lower values mean lower latencies for initial faults, but at the same time 671extra faults and I/O delays for following faults if they would have been part of 672that consecutive pages readahead would have brought in. 673 674============================================================= 675 676panic_on_oom 677 678This enables or disables panic on out-of-memory feature. 679 680If this is set to 0, the kernel will kill some rogue process, 681called oom_killer. Usually, oom_killer can kill rogue processes and 682system will survive. 683 684If this is set to 1, the kernel panics when out-of-memory happens. 685However, if a process limits using nodes by mempolicy/cpusets, 686and those nodes become memory exhaustion status, one process 687may be killed by oom-killer. No panic occurs in this case. 688Because other nodes' memory may be free. This means system total status 689may be not fatal yet. 690 691If this is set to 2, the kernel panics compulsorily even on the 692above-mentioned. Even oom happens under memory cgroup, the whole 693system panics. 694 695The default value is 0. 6961 and 2 are for failover of clustering. Please select either 697according to your policy of failover. 698panic_on_oom=2+kdump gives you very strong tool to investigate 699why oom happens. You can get snapshot. 700 701============================================================= 702 703percpu_pagelist_fraction 704 705This is the fraction of pages at most (high mark pcp->high) in each zone that 706are allocated for each per cpu page list. The min value for this is 8. It 707means that we don't allow more than 1/8th of pages in each zone to be 708allocated in any single per_cpu_pagelist. This entry only changes the value 709of hot per cpu pagelists. User can specify a number like 100 to allocate 7101/100th of each zone to each per cpu page list. 711 712The batch value of each per cpu pagelist is also updated as a result. It is 713set to pcp->high/4. The upper limit of batch is (PAGE_SHIFT * 8) 714 715The initial value is zero. Kernel does not use this value at boot time to set 716the high water marks for each per cpu page list. If the user writes '0' to this 717sysctl, it will revert to this default behavior. 718 719============================================================== 720 721stat_interval 722 723The time interval between which vm statistics are updated. The default 724is 1 second. 725 726============================================================== 727 728swappiness 729 730This control is used to define how aggressive the kernel will swap 731memory pages. Higher values will increase agressiveness, lower values 732decrease the amount of swap. A value of 0 instructs the kernel not to 733initiate swap until the amount of free and file-backed pages is less 734than the high water mark in a zone. 735 736The default value is 60. 737 738============================================================== 739 740- user_reserve_kbytes 741 742When overcommit_memory is set to 2, "never overcommit" mode, reserve 743min(3% of current process size, user_reserve_kbytes) of free memory. 744This is intended to prevent a user from starting a single memory hogging 745process, such that they cannot recover (kill the hog). 746 747user_reserve_kbytes defaults to min(3% of the current process size, 128MB). 748 749If this is reduced to zero, then the user will be allowed to allocate 750all free memory with a single process, minus admin_reserve_kbytes. 751Any subsequent attempts to execute a command will result in 752"fork: Cannot allocate memory". 753 754Changing this takes effect whenever an application requests memory. 755 756============================================================== 757 758vfs_cache_pressure 759------------------ 760 761This percentage value controls the tendency of the kernel to reclaim 762the memory which is used for caching of directory and inode objects. 763 764At the default value of vfs_cache_pressure=100 the kernel will attempt to 765reclaim dentries and inodes at a "fair" rate with respect to pagecache and 766swapcache reclaim. Decreasing vfs_cache_pressure causes the kernel to prefer 767to retain dentry and inode caches. When vfs_cache_pressure=0, the kernel will 768never reclaim dentries and inodes due to memory pressure and this can easily 769lead to out-of-memory conditions. Increasing vfs_cache_pressure beyond 100 770causes the kernel to prefer to reclaim dentries and inodes. 771 772Increasing vfs_cache_pressure significantly beyond 100 may have negative 773performance impact. Reclaim code needs to take various locks to find freeable 774directory and inode objects. With vfs_cache_pressure=1000, it will look for 775ten times more freeable objects than there are. 776 777============================================================== 778 779zone_reclaim_mode: 780 781Zone_reclaim_mode allows someone to set more or less aggressive approaches to 782reclaim memory when a zone runs out of memory. If it is set to zero then no 783zone reclaim occurs. Allocations will be satisfied from other zones / nodes 784in the system. 785 786This is value ORed together of 787 7881 = Zone reclaim on 7892 = Zone reclaim writes dirty pages out 7904 = Zone reclaim swaps pages 791 792zone_reclaim_mode is disabled by default. For file servers or workloads 793that benefit from having their data cached, zone_reclaim_mode should be 794left disabled as the caching effect is likely to be more important than 795data locality. 796 797zone_reclaim may be enabled if it's known that the workload is partitioned 798such that each partition fits within a NUMA node and that accessing remote 799memory would cause a measurable performance reduction. The page allocator 800will then reclaim easily reusable pages (those page cache pages that are 801currently not used) before allocating off node pages. 802 803Allowing zone reclaim to write out pages stops processes that are 804writing large amounts of data from dirtying pages on other nodes. Zone 805reclaim will write out dirty pages if a zone fills up and so effectively 806throttle the process. This may decrease the performance of a single process 807since it cannot use all of system memory to buffer the outgoing writes 808anymore but it preserve the memory on other nodes so that the performance 809of other processes running on other nodes will not be affected. 810 811Allowing regular swap effectively restricts allocations to the local 812node unless explicitly overridden by memory policies or cpuset 813configurations. 814 815============ End of Document ================================= 816