linux/Documentation/sysctl/vm.txt
<<
>>
Prefs
   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- dirty_background_bytes
  25- dirty_background_ratio
  26- dirty_bytes
  27- dirty_expire_centisecs
  28- dirty_ratio
  29- dirty_writeback_centisecs
  30- drop_caches
  31- extfrag_threshold
  32- hugepages_treat_as_movable
  33- hugetlb_shm_group
  34- laptop_mode
  35- legacy_va_layout
  36- lowmem_reserve_ratio
  37- max_map_count
  38- memory_failure_early_kill
  39- memory_failure_recovery
  40- min_free_kbytes
  41- min_slab_ratio
  42- min_unmapped_ratio
  43- mmap_min_addr
  44- nr_hugepages
  45- nr_overcommit_hugepages
  46- nr_trim_pages         (only if CONFIG_MMU=n)
  47- numa_zonelist_order
  48- oom_dump_tasks
  49- oom_kill_allocating_task
  50- overcommit_memory
  51- overcommit_ratio
  52- page-cluster
  53- panic_on_oom
  54- percpu_pagelist_fraction
  55- stat_interval
  56- swappiness
  57- user_reserve_kbytes
  58- vfs_cache_pressure
  59- zone_reclaim_mode
  60
  61==============================================================
  62
  63admin_reserve_kbytes
  64
  65The amount of free memory in the system that should be reserved for users
  66with the capability cap_sys_admin.
  67
  68admin_reserve_kbytes defaults to min(3% of free pages, 8MB)
  69
  70That should provide enough for the admin to log in and kill a process,
  71if necessary, under the default overcommit 'guess' mode.
  72
  73Systems running under overcommit 'never' should increase this to account
  74for the full Virtual Memory Size of programs used to recover. Otherwise,
  75root may not be able to log in to recover the system.
  76
  77How do you calculate a minimum useful reserve?
  78
  79sshd or login + bash (or some other shell) + top (or ps, kill, etc.)
  80
  81For overcommit 'guess', we can sum resident set sizes (RSS).
  82On x86_64 this is about 8MB.
  83
  84For overcommit 'never', we can take the max of their virtual sizes (VSZ)
  85and add the sum of their RSS.
  86On x86_64 this is about 128MB.
  87
  88Changing this takes effect whenever an application requests memory.
  89
  90==============================================================
  91
  92block_dump
  93
  94block_dump enables block I/O debugging when set to a nonzero value. More
  95information on block I/O debugging is in Documentation/laptops/laptop-mode.txt.
  96
  97==============================================================
  98
  99compact_memory
 100
 101Available only when CONFIG_COMPACTION is set. When 1 is written to the file,
 102all zones are compacted such that free memory is available in contiguous
 103blocks where possible. This can be important for example in the allocation of
 104huge pages although processes will also directly compact memory as required.
 105
 106==============================================================
 107
 108dirty_background_bytes
 109
 110Contains the amount of dirty memory at which the background kernel
 111flusher threads will start writeback.
 112
 113Note: dirty_background_bytes is the counterpart of dirty_background_ratio. Only
 114one of them may be specified at a time. When one sysctl is written it is
 115immediately taken into account to evaluate the dirty memory limits and the
 116other appears as 0 when read.
 117
 118==============================================================
 119
 120dirty_background_ratio
 121
 122Contains, as a percentage of total system memory, the number of pages at which
 123the background kernel flusher threads will start writing out dirty data.
 124
 125==============================================================
 126
 127dirty_bytes
 128
 129Contains the amount of dirty memory at which a process generating disk writes
 130will itself start writeback.
 131
 132Note: dirty_bytes is the counterpart of dirty_ratio. Only one of them may be
 133specified at a time. When one sysctl is written it is immediately taken into
 134account to evaluate the dirty memory limits and the other appears as 0 when
 135read.
 136
 137Note: the minimum value allowed for dirty_bytes is two pages (in bytes); any
 138value lower than this limit will be ignored and the old configuration will be
 139retained.
 140
 141==============================================================
 142
 143dirty_expire_centisecs
 144
 145This tunable is used to define when dirty data is old enough to be eligible
 146for writeout by the kernel flusher threads.  It is expressed in 100'ths
 147of a second.  Data which has been dirty in-memory for longer than this
 148interval will be written out next time a flusher thread wakes up.
 149
 150==============================================================
 151
 152dirty_ratio
 153
 154Contains, as a percentage of total system memory, the number of pages at which
 155a process which is generating disk writes will itself start writing out dirty
 156data.
 157
 158==============================================================
 159
 160dirty_writeback_centisecs
 161
 162The kernel flusher threads will periodically wake up and write `old' data
 163out to disk.  This tunable expresses the interval between those wakeups, in
 164100'ths of a second.
 165
 166Setting this to zero disables periodic writeback altogether.
 167
 168==============================================================
 169
 170drop_caches
 171
 172Writing to this will cause the kernel to drop clean caches, dentries and
 173inodes from memory, causing that memory to become free.
 174
 175To free pagecache:
 176        echo 1 > /proc/sys/vm/drop_caches
 177To free dentries and inodes:
 178        echo 2 > /proc/sys/vm/drop_caches
 179To free pagecache, dentries and inodes:
 180        echo 3 > /proc/sys/vm/drop_caches
 181
 182As this is a non-destructive operation and dirty objects are not freeable, the
 183user should run `sync' first.
 184
 185==============================================================
 186
 187extfrag_threshold
 188
 189This parameter affects whether the kernel will compact memory or direct
 190reclaim to satisfy a high-order allocation. /proc/extfrag_index shows what
 191the fragmentation index for each order is in each zone in the system. Values
 192tending towards 0 imply allocations would fail due to lack of memory,
 193values towards 1000 imply failures are due to fragmentation and -1 implies
 194that the allocation will succeed as long as watermarks are met.
 195
 196The kernel will not compact memory in a zone if the
 197fragmentation index is <= extfrag_threshold. The default value is 500.
 198
 199==============================================================
 200
 201hugepages_treat_as_movable
 202
 203This parameter is only useful when kernelcore= is specified at boot time to
 204create ZONE_MOVABLE for pages that may be reclaimed or migrated. Huge pages
 205are not movable so are not normally allocated from ZONE_MOVABLE. A non-zero
 206value written to hugepages_treat_as_movable allows huge pages to be allocated
 207from ZONE_MOVABLE.
 208
 209Once enabled, the ZONE_MOVABLE is treated as an area of memory the huge
 210pages pool can easily grow or shrink within. Assuming that applications are
 211not running that mlock() a lot of memory, it is likely the huge pages pool
 212can grow to the size of ZONE_MOVABLE by repeatedly entering the desired value
 213into nr_hugepages and triggering page reclaim.
 214
 215==============================================================
 216
 217hugetlb_shm_group
 218
 219hugetlb_shm_group contains group id that is allowed to create SysV
 220shared memory segment using hugetlb page.
 221
 222==============================================================
 223
 224laptop_mode
 225
 226laptop_mode is a knob that controls "laptop mode". All the things that are
 227controlled by this knob are discussed in Documentation/laptops/laptop-mode.txt.
 228
 229==============================================================
 230
 231legacy_va_layout
 232
 233If non-zero, this sysctl disables the new 32-bit mmap layout - the kernel
 234will use the legacy (2.4) layout for all processes.
 235
 236==============================================================
 237
 238lowmem_reserve_ratio
 239
 240For some specialised workloads on highmem machines it is dangerous for
 241the kernel to allow process memory to be allocated from the "lowmem"
 242zone.  This is because that memory could then be pinned via the mlock()
 243system call, or by unavailability of swapspace.
 244
 245And on large highmem machines this lack of reclaimable lowmem memory
 246can be fatal.
 247
 248So the Linux page allocator has a mechanism which prevents allocations
 249which _could_ use highmem from using too much lowmem.  This means that
 250a certain amount of lowmem is defended from the possibility of being
 251captured into pinned user memory.
 252
 253(The same argument applies to the old 16 megabyte ISA DMA region.  This
 254mechanism will also defend that region from allocations which could use
 255highmem or lowmem).
 256
 257The `lowmem_reserve_ratio' tunable determines how aggressive the kernel is
 258in defending these lower zones.
 259
 260If you have a machine which uses highmem or ISA DMA and your
 261applications are using mlock(), or if you are running with no swap then
 262you probably should change the lowmem_reserve_ratio setting.
 263
 264The lowmem_reserve_ratio is an array. You can see them by reading this file.
 265-
 266% cat /proc/sys/vm/lowmem_reserve_ratio
 267256     256     32
 268-
 269Note: # of this elements is one fewer than number of zones. Because the highest
 270      zone's value is not necessary for following calculation.
 271
 272But, these values are not used directly. The kernel calculates # of protection
 273pages for each zones from them. These are shown as array of protection pages
 274in /proc/zoneinfo like followings. (This is an example of x86-64 box).
 275Each zone has an array of protection pages like this.
 276
 277-
 278Node 0, zone      DMA
 279  pages free     1355
 280        min      3
 281        low      3
 282        high     4
 283        :
 284        :
 285    numa_other   0
 286        protection: (0, 2004, 2004, 2004)
 287        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 288  pagesets
 289    cpu: 0 pcp: 0
 290        :
 291-
 292These protections are added to score to judge whether this zone should be used
 293for page allocation or should be reclaimed.
 294
 295In this example, if normal pages (index=2) are required to this DMA zone and
 296watermark[WMARK_HIGH] is used for watermark, the kernel judges this zone should
 297not be used because pages_free(1355) is smaller than watermark + protection[2]
 298(4 + 2004 = 2008). If this protection value is 0, this zone would be used for
 299normal page requirement. If requirement is DMA zone(index=0), protection[0]
 300(=0) is used.
 301
 302zone[i]'s protection[j] is calculated by following expression.
 303
 304(i < j):
 305  zone[i]->protection[j]
 306  = (total sums of present_pages from zone[i+1] to zone[j] on the node)
 307    / lowmem_reserve_ratio[i];
 308(i = j):
 309   (should not be protected. = 0;
 310(i > j):
 311   (not necessary, but looks 0)
 312
 313The default values of lowmem_reserve_ratio[i] are
 314    256 (if zone[i] means DMA or DMA32 zone)
 315    32  (others).
 316As above expression, they are reciprocal number of ratio.
 317256 means 1/256. # of protection pages becomes about "0.39%" of total present
 318pages of higher zones on the node.
 319
 320If you would like to protect more pages, smaller values are effective.
 321The minimum value is 1 (1/1 -> 100%).
 322
 323==============================================================
 324
 325max_map_count:
 326
 327This file contains the maximum number of memory map areas a process
 328may have. Memory map areas are used as a side-effect of calling
 329malloc, directly by mmap and mprotect, and also when loading shared
 330libraries.
 331
 332While most applications need less than a thousand maps, certain
 333programs, particularly malloc debuggers, may consume lots of them,
 334e.g., up to one or two maps per allocation.
 335
 336The default value is 65536.
 337
 338=============================================================
 339
 340memory_failure_early_kill:
 341
 342Control how to kill processes when uncorrected memory error (typically
 343a 2bit error in a memory module) is detected in the background by hardware
 344that cannot be handled by the kernel. In some cases (like the page
 345still having a valid copy on disk) the kernel will handle the failure
 346transparently without affecting any applications. But if there is
 347no other uptodate copy of the data it will kill to prevent any data
 348corruptions from propagating.
 349
 3501: Kill all processes that have the corrupted and not reloadable page mapped
 351as soon as the corruption is detected.  Note this is not supported
 352for a few types of pages, like kernel internally allocated data or
 353the swap cache, but works for the majority of user pages.
 354
 3550: Only unmap the corrupted page from all processes and only kill a process
 356who tries to access it.
 357
 358The kill is done using a catchable SIGBUS with BUS_MCEERR_AO, so processes can
 359handle this if they want to.
 360
 361This is only active on architectures/platforms with advanced machine
 362check handling and depends on the hardware capabilities.
 363
 364Applications can override this setting individually with the PR_MCE_KILL prctl
 365
 366==============================================================
 367
 368memory_failure_recovery
 369
 370Enable memory failure recovery (when supported by the platform)
 371
 3721: Attempt recovery.
 373
 3740: Always panic on a memory failure.
 375
 376==============================================================
 377
 378min_free_kbytes:
 379
 380This is used to force the Linux VM to keep a minimum number
 381of kilobytes free.  The VM uses this number to compute a
 382watermark[WMARK_MIN] value for each lowmem zone in the system.
 383Each lowmem zone gets a number of reserved free pages based
 384proportionally on its size.
 385
 386Some minimal amount of memory is needed to satisfy PF_MEMALLOC
 387allocations; if you set this to lower than 1024KB, your system will
 388become subtly broken, and prone to deadlock under high loads.
 389
 390Setting this too high will OOM your machine instantly.
 391
 392=============================================================
 393
 394min_slab_ratio:
 395
 396This is available only on NUMA kernels.
 397
 398A percentage of the total pages in each zone.  On Zone reclaim
 399(fallback from the local zone occurs) slabs will be reclaimed if more
 400than this percentage of pages in a zone are reclaimable slab pages.
 401This insures that the slab growth stays under control even in NUMA
 402systems that rarely perform global reclaim.
 403
 404The default is 5 percent.
 405
 406Note that slab reclaim is triggered in a per zone / node fashion.
 407The process of reclaiming slab memory is currently not node specific
 408and may not be fast.
 409
 410=============================================================
 411
 412min_unmapped_ratio:
 413
 414This is available only on NUMA kernels.
 415
 416This is a percentage of the total pages in each zone. Zone reclaim will
 417only occur if more than this percentage of pages are in a state that
 418zone_reclaim_mode allows to be reclaimed.
 419
 420If zone_reclaim_mode has the value 4 OR'd, then the percentage is compared
 421against all file-backed unmapped pages including swapcache pages and tmpfs
 422files. Otherwise, only unmapped pages backed by normal files but not tmpfs
 423files and similar are considered.
 424
 425The default is 1 percent.
 426
 427==============================================================
 428
 429mmap_min_addr
 430
 431This file indicates the amount of address space  which a user process will
 432be restricted from mmapping.  Since kernel null dereference bugs could
 433accidentally operate based on the information in the first couple of pages
 434of memory userspace processes should not be allowed to write to them.  By
 435default this value is set to 0 and no protections will be enforced by the
 436security module.  Setting this value to something like 64k will allow the
 437vast majority of applications to work correctly and provide defense in depth
 438against future potential kernel bugs.
 439
 440==============================================================
 441
 442nr_hugepages
 443
 444Change the minimum size of the hugepage pool.
 445
 446See Documentation/vm/hugetlbpage.txt
 447
 448==============================================================
 449
 450nr_overcommit_hugepages
 451
 452Change the maximum size of the hugepage pool. The maximum is
 453nr_hugepages + nr_overcommit_hugepages.
 454
 455See Documentation/vm/hugetlbpage.txt
 456
 457==============================================================
 458
 459nr_trim_pages
 460
 461This is available only on NOMMU kernels.
 462
 463This value adjusts the excess page trimming behaviour of power-of-2 aligned
 464NOMMU mmap allocations.
 465
 466A value of 0 disables trimming of allocations entirely, while a value of 1
 467trims excess pages aggressively. Any value >= 1 acts as the watermark where
 468trimming of allocations is initiated.
 469
 470The default value is 1.
 471
 472See Documentation/nommu-mmap.txt for more information.
 473
 474==============================================================
 475
 476numa_zonelist_order
 477
 478This sysctl is only for NUMA.
 479'where the memory is allocated from' is controlled by zonelists.
 480(This documentation ignores ZONE_HIGHMEM/ZONE_DMA32 for simple explanation.
 481 you may be able to read ZONE_DMA as ZONE_DMA32...)
 482
 483In non-NUMA case, a zonelist for GFP_KERNEL is ordered as following.
 484ZONE_NORMAL -> ZONE_DMA
 485This means that a memory allocation request for GFP_KERNEL will
 486get memory from ZONE_DMA only when ZONE_NORMAL is not available.
 487
 488In NUMA case, you can think of following 2 types of order.
 489Assume 2 node NUMA and below is zonelist of Node(0)'s GFP_KERNEL
 490
 491(A) Node(0) ZONE_NORMAL -> Node(0) ZONE_DMA -> Node(1) ZONE_NORMAL
 492(B) Node(0) ZONE_NORMAL -> Node(1) ZONE_NORMAL -> Node(0) ZONE_DMA.
 493
 494Type(A) offers the best locality for processes on Node(0), but ZONE_DMA
 495will be used before ZONE_NORMAL exhaustion. This increases possibility of
 496out-of-memory(OOM) of ZONE_DMA because ZONE_DMA is tend to be small.
 497
 498Type(B) cannot offer the best locality but is more robust against OOM of
 499the DMA zone.
 500
 501Type(A) is called as "Node" order. Type (B) is "Zone" order.
 502
 503"Node order" orders the zonelists by node, then by zone within each node.
 504Specify "[Nn]ode" for node order
 505
 506"Zone Order" orders the zonelists by zone type, then by node within each
 507zone.  Specify "[Zz]one" for zone order.
 508
 509Specify "[Dd]efault" to request automatic configuration.  Autoconfiguration
 510will select "node" order in following case.
 511(1) if the DMA zone does not exist or
 512(2) if the DMA zone comprises greater than 50% of the available memory or
 513(3) if any node's DMA zone comprises greater than 60% of its local memory and
 514    the amount of local memory is big enough.
 515
 516Otherwise, "zone" order will be selected. Default order is recommended unless
 517this is causing problems for your system/application.
 518
 519==============================================================
 520
 521oom_dump_tasks
 522
 523Enables a system-wide task dump (excluding kernel threads) to be
 524produced when the kernel performs an OOM-killing and includes such
 525information as pid, uid, tgid, vm size, rss, nr_ptes, swapents,
 526oom_score_adj score, and name.  This is helpful to determine why the
 527OOM killer was invoked, to identify the rogue task that caused it,
 528and to determine why the OOM killer chose the task it did to kill.
 529
 530If this is set to zero, this information is suppressed.  On very
 531large systems with thousands of tasks it may not be feasible to dump
 532the memory state information for each one.  Such systems should not
 533be forced to incur a performance penalty in OOM conditions when the
 534information may not be desired.
 535
 536If this is set to non-zero, this information is shown whenever the
 537OOM killer actually kills a memory-hogging task.
 538
 539The default value is 1 (enabled).
 540
 541==============================================================
 542
 543oom_kill_allocating_task
 544
 545This enables or disables killing the OOM-triggering task in
 546out-of-memory situations.
 547
 548If this is set to zero, the OOM killer will scan through the entire
 549tasklist and select a task based on heuristics to kill.  This normally
 550selects a rogue memory-hogging task that frees up a large amount of
 551memory when killed.
 552
 553If this is set to non-zero, the OOM killer simply kills the task that
 554triggered the out-of-memory condition.  This avoids the expensive
 555tasklist scan.
 556
 557If panic_on_oom is selected, it takes precedence over whatever value
 558is used in oom_kill_allocating_task.
 559
 560The default value is 0.
 561
 562==============================================================
 563
 564overcommit_memory:
 565
 566This value contains a flag that enables memory overcommitment.
 567
 568When this flag is 0, the kernel attempts to estimate the amount
 569of free memory left when userspace requests more memory.
 570
 571When this flag is 1, the kernel pretends there is always enough
 572memory until it actually runs out.
 573
 574When this flag is 2, the kernel uses a "never overcommit"
 575policy that attempts to prevent any overcommit of memory.
 576Note that user_reserve_kbytes affects this policy.
 577
 578This feature can be very useful because there are a lot of
 579programs that malloc() huge amounts of memory "just-in-case"
 580and don't use much of it.
 581
 582The default value is 0.
 583
 584See Documentation/vm/overcommit-accounting and
 585security/commoncap.c::cap_vm_enough_memory() for more information.
 586
 587==============================================================
 588
 589overcommit_ratio:
 590
 591When overcommit_memory is set to 2, the committed address
 592space is not permitted to exceed swap plus this percentage
 593of physical RAM.  See above.
 594
 595==============================================================
 596
 597page-cluster
 598
 599page-cluster controls the number of pages up to which consecutive pages
 600are read in from swap in a single attempt. This is the swap counterpart
 601to page cache readahead.
 602The mentioned consecutivity is not in terms of virtual/physical addresses,
 603but consecutive on swap space - that means they were swapped out together.
 604
 605It is a logarithmic value - setting it to zero means "1 page", setting
 606it to 1 means "2 pages", setting it to 2 means "4 pages", etc.
 607Zero disables swap readahead completely.
 608
 609The default value is three (eight pages at a time).  There may be some
 610small benefits in tuning this to a different value if your workload is
 611swap-intensive.
 612
 613Lower values mean lower latencies for initial faults, but at the same time
 614extra faults and I/O delays for following faults if they would have been part of
 615that consecutive pages readahead would have brought in.
 616
 617=============================================================
 618
 619panic_on_oom
 620
 621This enables or disables panic on out-of-memory feature.
 622
 623If this is set to 0, the kernel will kill some rogue process,
 624called oom_killer.  Usually, oom_killer can kill rogue processes and
 625system will survive.
 626
 627If this is set to 1, the kernel panics when out-of-memory happens.
 628However, if a process limits using nodes by mempolicy/cpusets,
 629and those nodes become memory exhaustion status, one process
 630may be killed by oom-killer. No panic occurs in this case.
 631Because other nodes' memory may be free. This means system total status
 632may be not fatal yet.
 633
 634If this is set to 2, the kernel panics compulsorily even on the
 635above-mentioned. Even oom happens under memory cgroup, the whole
 636system panics.
 637
 638The default value is 0.
 6391 and 2 are for failover of clustering. Please select either
 640according to your policy of failover.
 641panic_on_oom=2+kdump gives you very strong tool to investigate
 642why oom happens. You can get snapshot.
 643
 644=============================================================
 645
 646percpu_pagelist_fraction
 647
 648This is the fraction of pages at most (high mark pcp->high) in each zone that
 649are allocated for each per cpu page list.  The min value for this is 8.  It
 650means that we don't allow more than 1/8th of pages in each zone to be
 651allocated in any single per_cpu_pagelist.  This entry only changes the value
 652of hot per cpu pagelists.  User can specify a number like 100 to allocate
 6531/100th of each zone to each per cpu page list.
 654
 655The batch value of each per cpu pagelist is also updated as a result.  It is
 656set to pcp->high/4.  The upper limit of batch is (PAGE_SHIFT * 8)
 657
 658The initial value is zero.  Kernel does not use this value at boot time to set
 659the high water marks for each per cpu page list.
 660
 661==============================================================
 662
 663stat_interval
 664
 665The time interval between which vm statistics are updated.  The default
 666is 1 second.
 667
 668==============================================================
 669
 670swappiness
 671
 672This control is used to define how aggressive the kernel will swap
 673memory pages.  Higher values will increase agressiveness, lower values
 674decrease the amount of swap.
 675
 676The default value is 60.
 677
 678==============================================================
 679
 680- user_reserve_kbytes
 681
 682When overcommit_memory is set to 2, "never overommit" mode, reserve
 683min(3% of current process size, user_reserve_kbytes) of free memory.
 684This is intended to prevent a user from starting a single memory hogging
 685process, such that they cannot recover (kill the hog).
 686
 687user_reserve_kbytes defaults to min(3% of the current process size, 128MB).
 688
 689If this is reduced to zero, then the user will be allowed to allocate
 690all free memory with a single process, minus admin_reserve_kbytes.
 691Any subsequent attempts to execute a command will result in
 692"fork: Cannot allocate memory".
 693
 694Changing this takes effect whenever an application requests memory.
 695
 696==============================================================
 697
 698vfs_cache_pressure
 699------------------
 700
 701Controls the tendency of the kernel to reclaim the memory which is used for
 702caching of directory and inode objects.
 703
 704At the default value of vfs_cache_pressure=100 the kernel will attempt to
 705reclaim dentries and inodes at a "fair" rate with respect to pagecache and
 706swapcache reclaim.  Decreasing vfs_cache_pressure causes the kernel to prefer
 707to retain dentry and inode caches. When vfs_cache_pressure=0, the kernel will
 708never reclaim dentries and inodes due to memory pressure and this can easily
 709lead to out-of-memory conditions. Increasing vfs_cache_pressure beyond 100
 710causes the kernel to prefer to reclaim dentries and inodes.
 711
 712==============================================================
 713
 714zone_reclaim_mode:
 715
 716Zone_reclaim_mode allows someone to set more or less aggressive approaches to
 717reclaim memory when a zone runs out of memory. If it is set to zero then no
 718zone reclaim occurs. Allocations will be satisfied from other zones / nodes
 719in the system.
 720
 721This is value ORed together of
 722
 7231       = Zone reclaim on
 7242       = Zone reclaim writes dirty pages out
 7254       = Zone reclaim swaps pages
 726
 727zone_reclaim_mode is set during bootup to 1 if it is determined that pages
 728from remote zones will cause a measurable performance reduction. The
 729page allocator will then reclaim easily reusable pages (those page
 730cache pages that are currently not used) before allocating off node pages.
 731
 732It may be beneficial to switch off zone reclaim if the system is
 733used for a file server and all of memory should be used for caching files
 734from disk. In that case the caching effect is more important than
 735data locality.
 736
 737Allowing zone reclaim to write out pages stops processes that are
 738writing large amounts of data from dirtying pages on other nodes. Zone
 739reclaim will write out dirty pages if a zone fills up and so effectively
 740throttle the process. This may decrease the performance of a single process
 741since it cannot use all of system memory to buffer the outgoing writes
 742anymore but it preserve the memory on other nodes so that the performance
 743of other processes running on other nodes will not be affected.
 744
 745Allowing regular swap effectively restricts allocations to the local
 746node unless explicitly overridden by memory policies or cpuset
 747configurations.
 748
 749============ End of Document =================================
 750
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.