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- 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- swappiness
  61- user_reserve_kbytes
  62- vfs_cache_pressure
  63- zone_reclaim_mode
  64
  65==============================================================
  66
  67admin_reserve_kbytes
  68
  69The amount of free memory in the system that should be reserved for users
  70with the capability cap_sys_admin.
  71
  72admin_reserve_kbytes defaults to min(3% of free pages, 8MB)
  73
  74That should provide enough for the admin to log in and kill a process,
  75if necessary, under the default overcommit 'guess' mode.
  76
  77Systems running under overcommit 'never' should increase this to account
  78for the full Virtual Memory Size of programs used to recover. Otherwise,
  79root may not be able to log in to recover the system.
  80
  81How do you calculate a minimum useful reserve?
  82
  83sshd or login + bash (or some other shell) + top (or ps, kill, etc.)
  84
  85For overcommit 'guess', we can sum resident set sizes (RSS).
  86On x86_64 this is about 8MB.
  87
  88For overcommit 'never', we can take the max of their virtual sizes (VSZ)
  89and add the sum of their RSS.
  90On x86_64 this is about 128MB.
  91
  92Changing this takes effect whenever an application requests memory.
  93
  94==============================================================
  95
  96block_dump
  97
  98block_dump enables block I/O debugging when set to a nonzero value. More
  99information on block I/O debugging is in Documentation/laptops/laptop-mode.txt.
 100
 101==============================================================
 102
 103compact_memory
 104
 105Available only when CONFIG_COMPACTION is set. When 1 is written to the file,
 106all zones are compacted such that free memory is available in contiguous
 107blocks where possible. This can be important for example in the allocation of
 108huge pages although processes will also directly compact memory as required.
 109
 110==============================================================
 111
 112compact_unevictable_allowed
 113
 114Available only when CONFIG_COMPACTION is set. When set to 1, compaction is
 115allowed to examine the unevictable lru (mlocked pages) for pages to compact.
 116This should be used on systems where stalls for minor page faults are an
 117acceptable trade for large contiguous free memory.  Set to 0 to prevent
 118compaction from moving pages that are unevictable.  Default value is 1.
 119
 120==============================================================
 121
 122dirty_background_bytes
 123
 124Contains the amount of dirty memory at which the background kernel
 125flusher threads will start writeback.
 126
 127Note: dirty_background_bytes is the counterpart of dirty_background_ratio. Only
 128one of them may be specified at a time. When one sysctl is written it is
 129immediately taken into account to evaluate the dirty memory limits and the
 130other appears as 0 when read.
 131
 132==============================================================
 133
 134dirty_background_ratio
 135
 136Contains, as a percentage of total available memory that contains free pages
 137and reclaimable pages, the number of pages at which the background kernel
 138flusher threads will start writing out dirty data.
 139
 140The total available memory is not equal to total system memory.
 141
 142==============================================================
 143
 144dirty_bytes
 145
 146Contains the amount of dirty memory at which a process generating disk writes
 147will itself start writeback.
 148
 149Note: dirty_bytes is the counterpart of dirty_ratio. Only one of them may be
 150specified at a time. When one sysctl is written it is immediately taken into
 151account to evaluate the dirty memory limits and the other appears as 0 when
 152read.
 153
 154Note: the minimum value allowed for dirty_bytes is two pages (in bytes); any
 155value lower than this limit will be ignored and the old configuration will be
 156retained.
 157
 158==============================================================
 159
 160dirty_expire_centisecs
 161
 162This tunable is used to define when dirty data is old enough to be eligible
 163for writeout by the kernel flusher threads.  It is expressed in 100'ths
 164of a second.  Data which has been dirty in-memory for longer than this
 165interval will be written out next time a flusher thread wakes up.
 166
 167==============================================================
 168
 169dirty_ratio
 170
 171Contains, as a percentage of total available memory that contains free pages
 172and reclaimable pages, the number of pages at which a process which is
 173generating disk writes will itself start writing out dirty data.
 174
 175The total available memory is not equal to total system memory.
 176
 177==============================================================
 178
 179dirty_writeback_centisecs
 180
 181The kernel flusher threads will periodically wake up and write `old' data
 182out to disk.  This tunable expresses the interval between those wakeups, in
 183100'ths of a second.
 184
 185Setting this to zero disables periodic writeback altogether.
 186
 187==============================================================
 188
 189drop_caches
 190
 191Writing to this will cause the kernel to drop clean caches, as well as
 192reclaimable slab objects like dentries and inodes.  Once dropped, their
 193memory becomes free.
 194
 195To free pagecache:
 196        echo 1 > /proc/sys/vm/drop_caches
 197To free reclaimable slab objects (includes dentries and inodes):
 198        echo 2 > /proc/sys/vm/drop_caches
 199To free slab objects and pagecache:
 200        echo 3 > /proc/sys/vm/drop_caches
 201
 202This is a non-destructive operation and will not free any dirty objects.
 203To increase the number of objects freed by this operation, the user may run
 204`sync' prior to writing to /proc/sys/vm/drop_caches.  This will minimize the
 205number of dirty objects on the system and create more candidates to be
 206dropped.
 207
 208This file is not a means to control the growth of the various kernel caches
 209(inodes, dentries, pagecache, etc...)  These objects are automatically
 210reclaimed by the kernel when memory is needed elsewhere on the system.
 211
 212Use of this file can cause performance problems.  Since it discards cached
 213objects, it may cost a significant amount of I/O and CPU to recreate the
 214dropped objects, especially if they were under heavy use.  Because of this,
 215use outside of a testing or debugging environment is not recommended.
 216
 217You may see informational messages in your kernel log when this file is
 218used:
 219
 220        cat (1234): drop_caches: 3
 221
 222These are informational only.  They do not mean that anything is wrong
 223with your system.  To disable them, echo 4 (bit 3) into drop_caches.
 224
 225==============================================================
 226
 227extfrag_threshold
 228
 229This parameter affects whether the kernel will compact memory or direct
 230reclaim to satisfy a high-order allocation. The extfrag/extfrag_index file in
 231debugfs shows what the fragmentation index for each order is in each zone in
 232the system. Values tending towards 0 imply allocations would fail due to lack
 233of memory, values towards 1000 imply failures are due to fragmentation and -1
 234implies that the allocation will succeed as long as watermarks are met.
 235
 236The kernel will not compact memory in a zone if the
 237fragmentation index is <= extfrag_threshold. The default value is 500.
 238
 239==============================================================
 240
 241hugepages_treat_as_movable
 242
 243This parameter controls whether we can allocate hugepages from ZONE_MOVABLE
 244or not. If set to non-zero, hugepages can be allocated from ZONE_MOVABLE.
 245ZONE_MOVABLE is created when kernel boot parameter kernelcore= is specified,
 246so this parameter has no effect if used without kernelcore=.
 247
 248Hugepage migration is now available in some situations which depend on the
 249architecture and/or the hugepage size. If a hugepage supports migration,
 250allocation from ZONE_MOVABLE is always enabled for the hugepage regardless
 251of the value of this parameter.
 252IOW, this parameter affects only non-migratable hugepages.
 253
 254Assuming that hugepages are not migratable in your system, one usecase of
 255this parameter is that users can make hugepage pool more extensible by
 256enabling the allocation from ZONE_MOVABLE. This is because on ZONE_MOVABLE
 257page reclaim/migration/compaction work more and you can get contiguous
 258memory more likely. Note that using ZONE_MOVABLE for non-migratable
 259hugepages can do harm to other features like memory hotremove (because
 260memory hotremove expects that memory blocks on ZONE_MOVABLE are always
 261removable,) so it's a trade-off responsible for the users.
 262
 263==============================================================
 264
 265hugetlb_shm_group
 266
 267hugetlb_shm_group contains group id that is allowed to create SysV
 268shared memory segment using hugetlb page.
 269
 270==============================================================
 271
 272laptop_mode
 273
 274laptop_mode is a knob that controls "laptop mode". All the things that are
 275controlled by this knob are discussed in Documentation/laptops/laptop-mode.txt.
 276
 277==============================================================
 278
 279legacy_va_layout
 280
 281If non-zero, this sysctl disables the new 32-bit mmap layout - the kernel
 282will use the legacy (2.4) layout for all processes.
 283
 284==============================================================
 285
 286lowmem_reserve_ratio
 287
 288For some specialised workloads on highmem machines it is dangerous for
 289the kernel to allow process memory to be allocated from the "lowmem"
 290zone.  This is because that memory could then be pinned via the mlock()
 291system call, or by unavailability of swapspace.
 292
 293And on large highmem machines this lack of reclaimable lowmem memory
 294can be fatal.
 295
 296So the Linux page allocator has a mechanism which prevents allocations
 297which _could_ use highmem from using too much lowmem.  This means that
 298a certain amount of lowmem is defended from the possibility of being
 299captured into pinned user memory.
 300
 301(The same argument applies to the old 16 megabyte ISA DMA region.  This
 302mechanism will also defend that region from allocations which could use
 303highmem or lowmem).
 304
 305The `lowmem_reserve_ratio' tunable determines how aggressive the kernel is
 306in defending these lower zones.
 307
 308If you have a machine which uses highmem or ISA DMA and your
 309applications are using mlock(), or if you are running with no swap then
 310you probably should change the lowmem_reserve_ratio setting.
 311
 312The lowmem_reserve_ratio is an array. You can see them by reading this file.
 313-
 314% cat /proc/sys/vm/lowmem_reserve_ratio
 315256     256     32
 316-
 317Note: # of this elements is one fewer than number of zones. Because the highest
 318      zone's value is not necessary for following calculation.
 319
 320But, these values are not used directly. The kernel calculates # of protection
 321pages for each zones from them. These are shown as array of protection pages
 322in /proc/zoneinfo like followings. (This is an example of x86-64 box).
 323Each zone has an array of protection pages like this.
 324
 325-
 326Node 0, zone      DMA
 327  pages free     1355
 328        min      3
 329        low      3
 330        high     4
 331        :
 332        :
 333    numa_other   0
 334        protection: (0, 2004, 2004, 2004)
 335        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 336  pagesets
 337    cpu: 0 pcp: 0
 338        :
 339-
 340These protections are added to score to judge whether this zone should be used
 341for page allocation or should be reclaimed.
 342
 343In this example, if normal pages (index=2) are required to this DMA zone and
 344watermark[WMARK_HIGH] is used for watermark, the kernel judges this zone should
 345not be used because pages_free(1355) is smaller than watermark + protection[2]
 346(4 + 2004 = 2008). If this protection value is 0, this zone would be used for
 347normal page requirement. If requirement is DMA zone(index=0), protection[0]
 348(=0) is used.
 349
 350zone[i]'s protection[j] is calculated by following expression.
 351
 352(i < j):
 353  zone[i]->protection[j]
 354  = (total sums of managed_pages from zone[i+1] to zone[j] on the node)
 355    / lowmem_reserve_ratio[i];
 356(i = j):
 357   (should not be protected. = 0;
 358(i > j):
 359   (not necessary, but looks 0)
 360
 361The default values of lowmem_reserve_ratio[i] are
 362    256 (if zone[i] means DMA or DMA32 zone)
 363    32  (others).
 364As above expression, they are reciprocal number of ratio.
 365256 means 1/256. # of protection pages becomes about "0.39%" of total managed
 366pages of higher zones on the node.
 367
 368If you would like to protect more pages, smaller values are effective.
 369The minimum value is 1 (1/1 -> 100%).
 370
 371==============================================================
 372
 373max_map_count:
 374
 375This file contains the maximum number of memory map areas a process
 376may have. Memory map areas are used as a side-effect of calling
 377malloc, directly by mmap and mprotect, and also when loading shared
 378libraries.
 379
 380While most applications need less than a thousand maps, certain
 381programs, particularly malloc debuggers, may consume lots of them,
 382e.g., up to one or two maps per allocation.
 383
 384The default value is 65536.
 385
 386=============================================================
 387
 388memory_failure_early_kill:
 389
 390Control how to kill processes when uncorrected memory error (typically
 391a 2bit error in a memory module) is detected in the background by hardware
 392that cannot be handled by the kernel. In some cases (like the page
 393still having a valid copy on disk) the kernel will handle the failure
 394transparently without affecting any applications. But if there is
 395no other uptodate copy of the data it will kill to prevent any data
 396corruptions from propagating.
 397
 3981: Kill all processes that have the corrupted and not reloadable page mapped
 399as soon as the corruption is detected.  Note this is not supported
 400for a few types of pages, like kernel internally allocated data or
 401the swap cache, but works for the majority of user pages.
 402
 4030: Only unmap the corrupted page from all processes and only kill a process
 404who tries to access it.
 405
 406The kill is done using a catchable SIGBUS with BUS_MCEERR_AO, so processes can
 407handle this if they want to.
 408
 409This is only active on architectures/platforms with advanced machine
 410check handling and depends on the hardware capabilities.
 411
 412Applications can override this setting individually with the PR_MCE_KILL prctl
 413
 414==============================================================
 415
 416memory_failure_recovery
 417
 418Enable memory failure recovery (when supported by the platform)
 419
 4201: Attempt recovery.
 421
 4220: Always panic on a memory failure.
 423
 424==============================================================
 425
 426min_free_kbytes:
 427
 428This is used to force the Linux VM to keep a minimum number
 429of kilobytes free.  The VM uses this number to compute a
 430watermark[WMARK_MIN] value for each lowmem zone in the system.
 431Each lowmem zone gets a number of reserved free pages based
 432proportionally on its size.
 433
 434Some minimal amount of memory is needed to satisfy PF_MEMALLOC
 435allocations; if you set this to lower than 1024KB, your system will
 436become subtly broken, and prone to deadlock under high loads.
 437
 438Setting this too high will OOM your machine instantly.
 439
 440=============================================================
 441
 442min_slab_ratio:
 443
 444This is available only on NUMA kernels.
 445
 446A percentage of the total pages in each zone.  On Zone reclaim
 447(fallback from the local zone occurs) slabs will be reclaimed if more
 448than this percentage of pages in a zone are reclaimable slab pages.
 449This insures that the slab growth stays under control even in NUMA
 450systems that rarely perform global reclaim.
 451
 452The default is 5 percent.
 453
 454Note that slab reclaim is triggered in a per zone / node fashion.
 455The process of reclaiming slab memory is currently not node specific
 456and may not be fast.
 457
 458=============================================================
 459
 460min_unmapped_ratio:
 461
 462This is available only on NUMA kernels.
 463
 464This is a percentage of the total pages in each zone. Zone reclaim will
 465only occur if more than this percentage of pages are in a state that
 466zone_reclaim_mode allows to be reclaimed.
 467
 468If zone_reclaim_mode has the value 4 OR'd, then the percentage is compared
 469against all file-backed unmapped pages including swapcache pages and tmpfs
 470files. Otherwise, only unmapped pages backed by normal files but not tmpfs
 471files and similar are considered.
 472
 473The default is 1 percent.
 474
 475==============================================================
 476
 477mmap_min_addr
 478
 479This file indicates the amount of address space  which a user process will
 480be restricted from mmapping.  Since kernel null dereference bugs could
 481accidentally operate based on the information in the first couple of pages
 482of memory userspace processes should not be allowed to write to them.  By
 483default this value is set to 0 and no protections will be enforced by the
 484security module.  Setting this value to something like 64k will allow the
 485vast majority of applications to work correctly and provide defense in depth
 486against future potential kernel bugs.
 487
 488==============================================================
 489
 490mmap_rnd_bits:
 491
 492This value can be used to select the number of bits to use to
 493determine the random offset to the base address of vma regions
 494resulting from mmap allocations on architectures which support
 495tuning address space randomization.  This value will be bounded
 496by the architecture's minimum and maximum supported values.
 497
 498This value can be changed after boot using the
 499/proc/sys/vm/mmap_rnd_bits tunable
 500
 501==============================================================
 502
 503mmap_rnd_compat_bits:
 504
 505This value can be used to select the number of bits to use to
 506determine the random offset to the base address of vma regions
 507resulting from mmap allocations for applications run in
 508compatibility mode on architectures which support tuning address
 509space randomization.  This value will be bounded by the
 510architecture's minimum and maximum supported values.
 511
 512This value can be changed after boot using the
 513/proc/sys/vm/mmap_rnd_compat_bits tunable
 514
 515==============================================================
 516
 517nr_hugepages
 518
 519Change the minimum size of the hugepage pool.
 520
 521See Documentation/vm/hugetlbpage.txt
 522
 523==============================================================
 524
 525nr_overcommit_hugepages
 526
 527Change the maximum size of the hugepage pool. The maximum is
 528nr_hugepages + nr_overcommit_hugepages.
 529
 530See Documentation/vm/hugetlbpage.txt
 531
 532==============================================================
 533
 534nr_trim_pages
 535
 536This is available only on NOMMU kernels.
 537
 538This value adjusts the excess page trimming behaviour of power-of-2 aligned
 539NOMMU mmap allocations.
 540
 541A value of 0 disables trimming of allocations entirely, while a value of 1
 542trims excess pages aggressively. Any value >= 1 acts as the watermark where
 543trimming of allocations is initiated.
 544
 545The default value is 1.
 546
 547See Documentation/nommu-mmap.txt for more information.
 548
 549==============================================================
 550
 551numa_zonelist_order
 552
 553This sysctl is only for NUMA.
 554'where the memory is allocated from' is controlled by zonelists.
 555(This documentation ignores ZONE_HIGHMEM/ZONE_DMA32 for simple explanation.
 556 you may be able to read ZONE_DMA as ZONE_DMA32...)
 557
 558In non-NUMA case, a zonelist for GFP_KERNEL is ordered as following.
 559ZONE_NORMAL -> ZONE_DMA
 560This means that a memory allocation request for GFP_KERNEL will
 561get memory from ZONE_DMA only when ZONE_NORMAL is not available.
 562
 563In NUMA case, you can think of following 2 types of order.
 564Assume 2 node NUMA and below is zonelist of Node(0)'s GFP_KERNEL
 565
 566(A) Node(0) ZONE_NORMAL -> Node(0) ZONE_DMA -> Node(1) ZONE_NORMAL
 567(B) Node(0) ZONE_NORMAL -> Node(1) ZONE_NORMAL -> Node(0) ZONE_DMA.
 568
 569Type(A) offers the best locality for processes on Node(0), but ZONE_DMA
 570will be used before ZONE_NORMAL exhaustion. This increases possibility of
 571out-of-memory(OOM) of ZONE_DMA because ZONE_DMA is tend to be small.
 572
 573Type(B) cannot offer the best locality but is more robust against OOM of
 574the DMA zone.
 575
 576Type(A) is called as "Node" order. Type (B) is "Zone" order.
 577
 578"Node order" orders the zonelists by node, then by zone within each node.
 579Specify "[Nn]ode" for node order
 580
 581"Zone Order" orders the zonelists by zone type, then by node within each
 582zone.  Specify "[Zz]one" for zone order.
 583
 584Specify "[Dd]efault" to request automatic configuration.
 585
 586On 32-bit, the Normal zone needs to be preserved for allocations accessible
 587by the kernel, so "zone" order will be selected.
 588
 589On 64-bit, devices that require DMA32/DMA are relatively rare, so "node"
 590order will be selected.
 591
 592Default order is recommended unless this is causing problems for your
 593system/application.
 594
 595==============================================================
 596
 597oom_dump_tasks
 598
 599Enables a system-wide task dump (excluding kernel threads) to be produced
 600when the kernel performs an OOM-killing and includes such information as
 601pid, uid, tgid, vm size, rss, nr_ptes, nr_pmds, swapents, oom_score_adj
 602score, and name.  This is helpful to determine why the OOM killer was
 603invoked, to identify the rogue task that caused it, and to determine why
 604the OOM killer chose the task it did to kill.
 605
 606If this is set to zero, this information is suppressed.  On very
 607large systems with thousands of tasks it may not be feasible to dump
 608the memory state information for each one.  Such systems should not
 609be forced to incur a performance penalty in OOM conditions when the
 610information may not be desired.
 611
 612If this is set to non-zero, this information is shown whenever the
 613OOM killer actually kills a memory-hogging task.
 614
 615The default value is 1 (enabled).
 616
 617==============================================================
 618
 619oom_kill_allocating_task
 620
 621This enables or disables killing the OOM-triggering task in
 622out-of-memory situations.
 623
 624If this is set to zero, the OOM killer will scan through the entire
 625tasklist and select a task based on heuristics to kill.  This normally
 626selects a rogue memory-hogging task that frees up a large amount of
 627memory when killed.
 628
 629If this is set to non-zero, the OOM killer simply kills the task that
 630triggered the out-of-memory condition.  This avoids the expensive
 631tasklist scan.
 632
 633If panic_on_oom is selected, it takes precedence over whatever value
 634is used in oom_kill_allocating_task.
 635
 636The default value is 0.
 637
 638==============================================================
 639
 640overcommit_kbytes:
 641
 642When overcommit_memory is set to 2, the committed address space is not
 643permitted to exceed swap plus this amount of physical RAM. See below.
 644
 645Note: overcommit_kbytes is the counterpart of overcommit_ratio. Only one
 646of them may be specified at a time. Setting one disables the other (which
 647then appears as 0 when read).
 648
 649==============================================================
 650
 651overcommit_memory:
 652
 653This value contains a flag that enables memory overcommitment.
 654
 655When this flag is 0, the kernel attempts to estimate the amount
 656of free memory left when userspace requests more memory.
 657
 658When this flag is 1, the kernel pretends there is always enough
 659memory until it actually runs out.
 660
 661When this flag is 2, the kernel uses a "never overcommit"
 662policy that attempts to prevent any overcommit of memory.
 663Note that user_reserve_kbytes affects this policy.
 664
 665This feature can be very useful because there are a lot of
 666programs that malloc() huge amounts of memory "just-in-case"
 667and don't use much of it.
 668
 669The default value is 0.
 670
 671See Documentation/vm/overcommit-accounting and
 672mm/mmap.c::__vm_enough_memory() for more information.
 673
 674==============================================================
 675
 676overcommit_ratio:
 677
 678When overcommit_memory is set to 2, the committed address
 679space is not permitted to exceed swap plus this percentage
 680of physical RAM.  See above.
 681
 682==============================================================
 683
 684page-cluster
 685
 686page-cluster controls the number of pages up to which consecutive pages
 687are read in from swap in a single attempt. This is the swap counterpart
 688to page cache readahead.
 689The mentioned consecutivity is not in terms of virtual/physical addresses,
 690but consecutive on swap space - that means they were swapped out together.
 691
 692It is a logarithmic value - setting it to zero means "1 page", setting
 693it to 1 means "2 pages", setting it to 2 means "4 pages", etc.
 694Zero disables swap readahead completely.
 695
 696The default value is three (eight pages at a time).  There may be some
 697small benefits in tuning this to a different value if your workload is
 698swap-intensive.
 699
 700Lower values mean lower latencies for initial faults, but at the same time
 701extra faults and I/O delays for following faults if they would have been part of
 702that consecutive pages readahead would have brought in.
 703
 704=============================================================
 705
 706panic_on_oom
 707
 708This enables or disables panic on out-of-memory feature.
 709
 710If this is set to 0, the kernel will kill some rogue process,
 711called oom_killer.  Usually, oom_killer can kill rogue processes and
 712system will survive.
 713
 714If this is set to 1, the kernel panics when out-of-memory happens.
 715However, if a process limits using nodes by mempolicy/cpusets,
 716and those nodes become memory exhaustion status, one process
 717may be killed by oom-killer. No panic occurs in this case.
 718Because other nodes' memory may be free. This means system total status
 719may be not fatal yet.
 720
 721If this is set to 2, the kernel panics compulsorily even on the
 722above-mentioned. Even oom happens under memory cgroup, the whole
 723system panics.
 724
 725The default value is 0.
 7261 and 2 are for failover of clustering. Please select either
 727according to your policy of failover.
 728panic_on_oom=2+kdump gives you very strong tool to investigate
 729why oom happens. You can get snapshot.
 730
 731=============================================================
 732
 733percpu_pagelist_fraction
 734
 735This is the fraction of pages at most (high mark pcp->high) in each zone that
 736are allocated for each per cpu page list.  The min value for this is 8.  It
 737means that we don't allow more than 1/8th of pages in each zone to be
 738allocated in any single per_cpu_pagelist.  This entry only changes the value
 739of hot per cpu pagelists.  User can specify a number like 100 to allocate
 7401/100th of each zone to each per cpu page list.
 741
 742The batch value of each per cpu pagelist is also updated as a result.  It is
 743set to pcp->high/4.  The upper limit of batch is (PAGE_SHIFT * 8)
 744
 745The initial value is zero.  Kernel does not use this value at boot time to set
 746the high water marks for each per cpu page list.  If the user writes '0' to this
 747sysctl, it will revert to this default behavior.
 748
 749==============================================================
 750
 751stat_interval
 752
 753The time interval between which vm statistics are updated.  The default
 754is 1 second.
 755
 756==============================================================
 757
 758swappiness
 759
 760This control is used to define how aggressive the kernel will swap
 761memory pages.  Higher values will increase agressiveness, lower values
 762decrease the amount of swap.  A value of 0 instructs the kernel not to
 763initiate swap until the amount of free and file-backed pages is less
 764than the high water mark in a zone.
 765
 766The default value is 60.
 767
 768==============================================================
 769
 770- user_reserve_kbytes
 771
 772When overcommit_memory is set to 2, "never overcommit" mode, reserve
 773min(3% of current process size, user_reserve_kbytes) of free memory.
 774This is intended to prevent a user from starting a single memory hogging
 775process, such that they cannot recover (kill the hog).
 776
 777user_reserve_kbytes defaults to min(3% of the current process size, 128MB).
 778
 779If this is reduced to zero, then the user will be allowed to allocate
 780all free memory with a single process, minus admin_reserve_kbytes.
 781Any subsequent attempts to execute a command will result in
 782"fork: Cannot allocate memory".
 783
 784Changing this takes effect whenever an application requests memory.
 785
 786==============================================================
 787
 788vfs_cache_pressure
 789------------------
 790
 791This percentage value controls the tendency of the kernel to reclaim
 792the memory which is used for caching of directory and inode objects.
 793
 794At the default value of vfs_cache_pressure=100 the kernel will attempt to
 795reclaim dentries and inodes at a "fair" rate with respect to pagecache and
 796swapcache reclaim.  Decreasing vfs_cache_pressure causes the kernel to prefer
 797to retain dentry and inode caches. When vfs_cache_pressure=0, the kernel will
 798never reclaim dentries and inodes due to memory pressure and this can easily
 799lead to out-of-memory conditions. Increasing vfs_cache_pressure beyond 100
 800causes the kernel to prefer to reclaim dentries and inodes.
 801
 802Increasing vfs_cache_pressure significantly beyond 100 may have negative
 803performance impact. Reclaim code needs to take various locks to find freeable
 804directory and inode objects. With vfs_cache_pressure=1000, it will look for
 805ten times more freeable objects than there are.
 806
 807=============================================================
 808
 809watermark_scale_factor:
 810
 811This factor controls the aggressiveness of kswapd. It defines the
 812amount of memory left in a node/system before kswapd is woken up and
 813how much memory needs to be free before kswapd goes back to sleep.
 814
 815The unit is in fractions of 10,000. The default value of 10 means the
 816distances between watermarks are 0.1% of the available memory in the
 817node/system. The maximum value is 1000, or 10% of memory.
 818
 819A high rate of threads entering direct reclaim (allocstall) or kswapd
 820going to sleep prematurely (kswapd_low_wmark_hit_quickly) can indicate
 821that the number of free pages kswapd maintains for latency reasons is
 822too small for the allocation bursts occurring in the system. This knob
 823can then be used to tune kswapd aggressiveness accordingly.
 824
 825==============================================================
 826
 827zone_reclaim_mode:
 828
 829Zone_reclaim_mode allows someone to set more or less aggressive approaches to
 830reclaim memory when a zone runs out of memory. If it is set to zero then no
 831zone reclaim occurs. Allocations will be satisfied from other zones / nodes
 832in the system.
 833
 834This is value ORed together of
 835
 8361       = Zone reclaim on
 8372       = Zone reclaim writes dirty pages out
 8384       = Zone reclaim swaps pages
 839
 840zone_reclaim_mode is disabled by default.  For file servers or workloads
 841that benefit from having their data cached, zone_reclaim_mode should be
 842left disabled as the caching effect is likely to be more important than
 843data locality.
 844
 845zone_reclaim may be enabled if it's known that the workload is partitioned
 846such that each partition fits within a NUMA node and that accessing remote
 847memory would cause a measurable performance reduction.  The page allocator
 848will then reclaim easily reusable pages (those page cache pages that are
 849currently not used) before allocating off node pages.
 850
 851Allowing zone reclaim to write out pages stops processes that are
 852writing large amounts of data from dirtying pages on other nodes. Zone
 853reclaim will write out dirty pages if a zone fills up and so effectively
 854throttle the process. This may decrease the performance of a single process
 855since it cannot use all of system memory to buffer the outgoing writes
 856anymore but it preserve the memory on other nodes so that the performance
 857of other processes running on other nodes will not be affected.
 858
 859Allowing regular swap effectively restricts allocations to the local
 860node unless explicitly overridden by memory policies or cpuset
 861configurations.
 862
 863============ End of Document =================================
 864
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.