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_kbytes
  51- overcommit_memory
  52- overcommit_ratio
  53- page-cluster
  54- panic_on_oom
  55- percpu_pagelist_fraction
  56- stat_interval
  57- swappiness
  58- user_reserve_kbytes
  59- vfs_cache_pressure
  60- zone_reclaim_mode
  61
  62==============================================================
  63
  64admin_reserve_kbytes
  65
  66The amount of free memory in the system that should be reserved for users
  67with the capability cap_sys_admin.
  68
  69admin_reserve_kbytes defaults to min(3% of free pages, 8MB)
  70
  71That should provide enough for the admin to log in and kill a process,
  72if necessary, under the default overcommit 'guess' mode.
  73
  74Systems running under overcommit 'never' should increase this to account
  75for the full Virtual Memory Size of programs used to recover. Otherwise,
  76root may not be able to log in to recover the system.
  77
  78How do you calculate a minimum useful reserve?
  79
  80sshd or login + bash (or some other shell) + top (or ps, kill, etc.)
  81
  82For overcommit 'guess', we can sum resident set sizes (RSS).
  83On x86_64 this is about 8MB.
  84
  85For overcommit 'never', we can take the max of their virtual sizes (VSZ)
  86and add the sum of their RSS.
  87On x86_64 this is about 128MB.
  88
  89Changing this takes effect whenever an application requests memory.
  90
  91==============================================================
  92
  93block_dump
  94
  95block_dump enables block I/O debugging when set to a nonzero value. More
  96information on block I/O debugging is in Documentation/laptops/laptop-mode.txt.
  97
  98==============================================================
  99
 100compact_memory
 101
 102Available only when CONFIG_COMPACTION is set. When 1 is written to the file,
 103all zones are compacted such that free memory is available in contiguous
 104blocks where possible. This can be important for example in the allocation of
 105huge pages although processes will also directly compact memory as required.
 106
 107==============================================================
 108
 109dirty_background_bytes
 110
 111Contains the amount of dirty memory at which the background kernel
 112flusher threads will start writeback.
 113
 114Note: dirty_background_bytes is the counterpart of dirty_background_ratio. Only
 115one of them may be specified at a time. When one sysctl is written it is
 116immediately taken into account to evaluate the dirty memory limits and the
 117other appears as 0 when read.
 118
 119==============================================================
 120
 121dirty_background_ratio
 122
 123Contains, as a percentage of total available memory that contains free pages
 124and reclaimable pages, the number of pages at which the background kernel
 125flusher threads will start writing out dirty data.
 126
 127The total avaiable memory is not equal to total system memory.
 128
 129==============================================================
 130
 131dirty_bytes
 132
 133Contains the amount of dirty memory at which a process generating disk writes
 134will itself start writeback.
 135
 136Note: dirty_bytes is the counterpart of dirty_ratio. Only one of them may be
 137specified at a time. When one sysctl is written it is immediately taken into
 138account to evaluate the dirty memory limits and the other appears as 0 when
 139read.
 140
 141Note: the minimum value allowed for dirty_bytes is two pages (in bytes); any
 142value lower than this limit will be ignored and the old configuration will be
 143retained.
 144
 145==============================================================
 146
 147dirty_expire_centisecs
 148
 149This tunable is used to define when dirty data is old enough to be eligible
 150for writeout by the kernel flusher threads.  It is expressed in 100'ths
 151of a second.  Data which has been dirty in-memory for longer than this
 152interval will be written out next time a flusher thread wakes up.
 153
 154==============================================================
 155
 156dirty_ratio
 157
 158Contains, as a percentage of total available memory that contains free pages
 159and reclaimable pages, the number of pages at which a process which is
 160generating disk writes will itself start writing out dirty data.
 161
 162The total avaiable memory is not equal to total system memory.
 163
 164==============================================================
 165
 166dirty_writeback_centisecs
 167
 168The kernel flusher threads will periodically wake up and write `old' data
 169out to disk.  This tunable expresses the interval between those wakeups, in
 170100'ths of a second.
 171
 172Setting this to zero disables periodic writeback altogether.
 173
 174==============================================================
 175
 176drop_caches
 177
 178Writing to this will cause the kernel to drop clean caches, as well as
 179reclaimable slab objects like dentries and inodes.  Once dropped, their
 180memory becomes free.
 181
 182To free pagecache:
 183        echo 1 > /proc/sys/vm/drop_caches
 184To free reclaimable slab objects (includes dentries and inodes):
 185        echo 2 > /proc/sys/vm/drop_caches
 186To free slab objects and pagecache:
 187        echo 3 > /proc/sys/vm/drop_caches
 188
 189This is a non-destructive operation and will not free any dirty objects.
 190To increase the number of objects freed by this operation, the user may run
 191`sync' prior to writing to /proc/sys/vm/drop_caches.  This will minimize the
 192number of dirty objects on the system and create more candidates to be
 193dropped.
 194
 195This file is not a means to control the growth of the various kernel caches
 196(inodes, dentries, pagecache, etc...)  These objects are automatically
 197reclaimed by the kernel when memory is needed elsewhere on the system.
 198
 199Use of this file can cause performance problems.  Since it discards cached
 200objects, it may cost a significant amount of I/O and CPU to recreate the
 201dropped objects, especially if they were under heavy use.  Because of this,
 202use outside of a testing or debugging environment is not recommended.
 203
 204You may see informational messages in your kernel log when this file is
 205used:
 206
 207        cat (1234): drop_caches: 3
 208
 209These are informational only.  They do not mean that anything is wrong
 210with your system.  To disable them, echo 4 (bit 3) into drop_caches.
 211
 212==============================================================
 213
 214extfrag_threshold
 215
 216This parameter affects whether the kernel will compact memory or direct
 217reclaim to satisfy a high-order allocation. /proc/extfrag_index shows what
 218the fragmentation index for each order is in each zone in the system. Values
 219tending towards 0 imply allocations would fail due to lack of memory,
 220values towards 1000 imply failures are due to fragmentation and -1 implies
 221that the allocation will succeed as long as watermarks are met.
 222
 223The kernel will not compact memory in a zone if the
 224fragmentation index is <= extfrag_threshold. The default value is 500.
 225
 226==============================================================
 227
 228hugepages_treat_as_movable
 229
 230This parameter controls whether we can allocate hugepages from ZONE_MOVABLE
 231or not. If set to non-zero, hugepages can be allocated from ZONE_MOVABLE.
 232ZONE_MOVABLE is created when kernel boot parameter kernelcore= is specified,
 233so this parameter has no effect if used without kernelcore=.
 234
 235Hugepage migration is now available in some situations which depend on the
 236architecture and/or the hugepage size. If a hugepage supports migration,
 237allocation from ZONE_MOVABLE is always enabled for the hugepage regardless
 238of the value of this parameter.
 239IOW, this parameter affects only non-migratable hugepages.
 240
 241Assuming that hugepages are not migratable in your system, one usecase of
 242this parameter is that users can make hugepage pool more extensible by
 243enabling the allocation from ZONE_MOVABLE. This is because on ZONE_MOVABLE
 244page reclaim/migration/compaction work more and you can get contiguous
 245memory more likely. Note that using ZONE_MOVABLE for non-migratable
 246hugepages can do harm to other features like memory hotremove (because
 247memory hotremove expects that memory blocks on ZONE_MOVABLE are always
 248removable,) so it's a trade-off responsible for the users.
 249
 250==============================================================
 251
 252hugetlb_shm_group
 253
 254hugetlb_shm_group contains group id that is allowed to create SysV
 255shared memory segment using hugetlb page.
 256
 257==============================================================
 258
 259laptop_mode
 260
 261laptop_mode is a knob that controls "laptop mode". All the things that are
 262controlled by this knob are discussed in Documentation/laptops/laptop-mode.txt.
 263
 264==============================================================
 265
 266legacy_va_layout
 267
 268If non-zero, this sysctl disables the new 32-bit mmap layout - the kernel
 269will use the legacy (2.4) layout for all processes.
 270
 271==============================================================
 272
 273lowmem_reserve_ratio
 274
 275For some specialised workloads on highmem machines it is dangerous for
 276the kernel to allow process memory to be allocated from the "lowmem"
 277zone.  This is because that memory could then be pinned via the mlock()
 278system call, or by unavailability of swapspace.
 279
 280And on large highmem machines this lack of reclaimable lowmem memory
 281can be fatal.
 282
 283So the Linux page allocator has a mechanism which prevents allocations
 284which _could_ use highmem from using too much lowmem.  This means that
 285a certain amount of lowmem is defended from the possibility of being
 286captured into pinned user memory.
 287
 288(The same argument applies to the old 16 megabyte ISA DMA region.  This
 289mechanism will also defend that region from allocations which could use
 290highmem or lowmem).
 291
 292The `lowmem_reserve_ratio' tunable determines how aggressive the kernel is
 293in defending these lower zones.
 294
 295If you have a machine which uses highmem or ISA DMA and your
 296applications are using mlock(), or if you are running with no swap then
 297you probably should change the lowmem_reserve_ratio setting.
 298
 299The lowmem_reserve_ratio is an array. You can see them by reading this file.
 300-
 301% cat /proc/sys/vm/lowmem_reserve_ratio
 302256     256     32
 303-
 304Note: # of this elements is one fewer than number of zones. Because the highest
 305      zone's value is not necessary for following calculation.
 306
 307But, these values are not used directly. The kernel calculates # of protection
 308pages for each zones from them. These are shown as array of protection pages
 309in /proc/zoneinfo like followings. (This is an example of x86-64 box).
 310Each zone has an array of protection pages like this.
 311
 312-
 313Node 0, zone      DMA
 314  pages free     1355
 315        min      3
 316        low      3
 317        high     4
 318        :
 319        :
 320    numa_other   0
 321        protection: (0, 2004, 2004, 2004)
 322        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 323  pagesets
 324    cpu: 0 pcp: 0
 325        :
 326-
 327These protections are added to score to judge whether this zone should be used
 328for page allocation or should be reclaimed.
 329
 330In this example, if normal pages (index=2) are required to this DMA zone and
 331watermark[WMARK_HIGH] is used for watermark, the kernel judges this zone should
 332not be used because pages_free(1355) is smaller than watermark + protection[2]
 333(4 + 2004 = 2008). If this protection value is 0, this zone would be used for
 334normal page requirement. If requirement is DMA zone(index=0), protection[0]
 335(=0) is used.
 336
 337zone[i]'s protection[j] is calculated by following expression.
 338
 339(i < j):
 340  zone[i]->protection[j]
 341  = (total sums of present_pages from zone[i+1] to zone[j] on the node)
 342    / lowmem_reserve_ratio[i];
 343(i = j):
 344   (should not be protected. = 0;
 345(i > j):
 346   (not necessary, but looks 0)
 347
 348The default values of lowmem_reserve_ratio[i] are
 349    256 (if zone[i] means DMA or DMA32 zone)
 350    32  (others).
 351As above expression, they are reciprocal number of ratio.
 352256 means 1/256. # of protection pages becomes about "0.39%" of total present
 353pages of higher zones on the node.
 354
 355If you would like to protect more pages, smaller values are effective.
 356The minimum value is 1 (1/1 -> 100%).
 357
 358==============================================================
 359
 360max_map_count:
 361
 362This file contains the maximum number of memory map areas a process
 363may have. Memory map areas are used as a side-effect of calling
 364malloc, directly by mmap and mprotect, and also when loading shared
 365libraries.
 366
 367While most applications need less than a thousand maps, certain
 368programs, particularly malloc debuggers, may consume lots of them,
 369e.g., up to one or two maps per allocation.
 370
 371The default value is 65536.
 372
 373=============================================================
 374
 375memory_failure_early_kill:
 376
 377Control how to kill processes when uncorrected memory error (typically
 378a 2bit error in a memory module) is detected in the background by hardware
 379that cannot be handled by the kernel. In some cases (like the page
 380still having a valid copy on disk) the kernel will handle the failure
 381transparently without affecting any applications. But if there is
 382no other uptodate copy of the data it will kill to prevent any data
 383corruptions from propagating.
 384
 3851: Kill all processes that have the corrupted and not reloadable page mapped
 386as soon as the corruption is detected.  Note this is not supported
 387for a few types of pages, like kernel internally allocated data or
 388the swap cache, but works for the majority of user pages.
 389
 3900: Only unmap the corrupted page from all processes and only kill a process
 391who tries to access it.
 392
 393The kill is done using a catchable SIGBUS with BUS_MCEERR_AO, so processes can
 394handle this if they want to.
 395
 396This is only active on architectures/platforms with advanced machine
 397check handling and depends on the hardware capabilities.
 398
 399Applications can override this setting individually with the PR_MCE_KILL prctl
 400
 401==============================================================
 402
 403memory_failure_recovery
 404
 405Enable memory failure recovery (when supported by the platform)
 406
 4071: Attempt recovery.
 408
 4090: Always panic on a memory failure.
 410
 411==============================================================
 412
 413min_free_kbytes:
 414
 415This is used to force the Linux VM to keep a minimum number
 416of kilobytes free.  The VM uses this number to compute a
 417watermark[WMARK_MIN] value for each lowmem zone in the system.
 418Each lowmem zone gets a number of reserved free pages based
 419proportionally on its size.
 420
 421Some minimal amount of memory is needed to satisfy PF_MEMALLOC
 422allocations; if you set this to lower than 1024KB, your system will
 423become subtly broken, and prone to deadlock under high loads.
 424
 425Setting this too high will OOM your machine instantly.
 426
 427=============================================================
 428
 429min_slab_ratio:
 430
 431This is available only on NUMA kernels.
 432
 433A percentage of the total pages in each zone.  On Zone reclaim
 434(fallback from the local zone occurs) slabs will be reclaimed if more
 435than this percentage of pages in a zone are reclaimable slab pages.
 436This insures that the slab growth stays under control even in NUMA
 437systems that rarely perform global reclaim.
 438
 439The default is 5 percent.
 440
 441Note that slab reclaim is triggered in a per zone / node fashion.
 442The process of reclaiming slab memory is currently not node specific
 443and may not be fast.
 444
 445=============================================================
 446
 447min_unmapped_ratio:
 448
 449This is available only on NUMA kernels.
 450
 451This is a percentage of the total pages in each zone. Zone reclaim will
 452only occur if more than this percentage of pages are in a state that
 453zone_reclaim_mode allows to be reclaimed.
 454
 455If zone_reclaim_mode has the value 4 OR'd, then the percentage is compared
 456against all file-backed unmapped pages including swapcache pages and tmpfs
 457files. Otherwise, only unmapped pages backed by normal files but not tmpfs
 458files and similar are considered.
 459
 460The default is 1 percent.
 461
 462==============================================================
 463
 464mmap_min_addr
 465
 466This file indicates the amount of address space  which a user process will
 467be restricted from mmapping.  Since kernel null dereference bugs could
 468accidentally operate based on the information in the first couple of pages
 469of memory userspace processes should not be allowed to write to them.  By
 470default this value is set to 0 and no protections will be enforced by the
 471security module.  Setting this value to something like 64k will allow the
 472vast majority of applications to work correctly and provide defense in depth
 473against future potential kernel bugs.
 474
 475==============================================================
 476
 477nr_hugepages
 478
 479Change the minimum size of the hugepage pool.
 480
 481See Documentation/vm/hugetlbpage.txt
 482
 483==============================================================
 484
 485nr_overcommit_hugepages
 486
 487Change the maximum size of the hugepage pool. The maximum is
 488nr_hugepages + nr_overcommit_hugepages.
 489
 490See Documentation/vm/hugetlbpage.txt
 491
 492==============================================================
 493
 494nr_trim_pages
 495
 496This is available only on NOMMU kernels.
 497
 498This value adjusts the excess page trimming behaviour of power-of-2 aligned
 499NOMMU mmap allocations.
 500
 501A value of 0 disables trimming of allocations entirely, while a value of 1
 502trims excess pages aggressively. Any value >= 1 acts as the watermark where
 503trimming of allocations is initiated.
 504
 505The default value is 1.
 506
 507See Documentation/nommu-mmap.txt for more information.
 508
 509==============================================================
 510
 511numa_zonelist_order
 512
 513This sysctl is only for NUMA.
 514'where the memory is allocated from' is controlled by zonelists.
 515(This documentation ignores ZONE_HIGHMEM/ZONE_DMA32 for simple explanation.
 516 you may be able to read ZONE_DMA as ZONE_DMA32...)
 517
 518In non-NUMA case, a zonelist for GFP_KERNEL is ordered as following.
 519ZONE_NORMAL -> ZONE_DMA
 520This means that a memory allocation request for GFP_KERNEL will
 521get memory from ZONE_DMA only when ZONE_NORMAL is not available.
 522
 523In NUMA case, you can think of following 2 types of order.
 524Assume 2 node NUMA and below is zonelist of Node(0)'s GFP_KERNEL
 525
 526(A) Node(0) ZONE_NORMAL -> Node(0) ZONE_DMA -> Node(1) ZONE_NORMAL
 527(B) Node(0) ZONE_NORMAL -> Node(1) ZONE_NORMAL -> Node(0) ZONE_DMA.
 528
 529Type(A) offers the best locality for processes on Node(0), but ZONE_DMA
 530will be used before ZONE_NORMAL exhaustion. This increases possibility of
 531out-of-memory(OOM) of ZONE_DMA because ZONE_DMA is tend to be small.
 532
 533Type(B) cannot offer the best locality but is more robust against OOM of
 534the DMA zone.
 535
 536Type(A) is called as "Node" order. Type (B) is "Zone" order.
 537
 538"Node order" orders the zonelists by node, then by zone within each node.
 539Specify "[Nn]ode" for node order
 540
 541"Zone Order" orders the zonelists by zone type, then by node within each
 542zone.  Specify "[Zz]one" for zone order.
 543
 544Specify "[Dd]efault" to request automatic configuration.  Autoconfiguration
 545will select "node" order in following case.
 546(1) if the DMA zone does not exist or
 547(2) if the DMA zone comprises greater than 50% of the available memory or
 548(3) if any node's DMA zone comprises greater than 70% of its local memory and
 549    the amount of local memory is big enough.
 550
 551Otherwise, "zone" order will be selected. Default order is recommended unless
 552this is causing problems for your system/application.
 553
 554==============================================================
 555
 556oom_dump_tasks
 557
 558Enables a system-wide task dump (excluding kernel threads) to be
 559produced when the kernel performs an OOM-killing and includes such
 560information as pid, uid, tgid, vm size, rss, nr_ptes, swapents,
 561oom_score_adj score, and name.  This is helpful to determine why the
 562OOM killer was invoked, to identify the rogue task that caused it,
 563and to determine why the OOM killer chose the task it did to kill.
 564
 565If this is set to zero, this information is suppressed.  On very
 566large systems with thousands of tasks it may not be feasible to dump
 567the memory state information for each one.  Such systems should not
 568be forced to incur a performance penalty in OOM conditions when the
 569information may not be desired.
 570
 571If this is set to non-zero, this information is shown whenever the
 572OOM killer actually kills a memory-hogging task.
 573
 574The default value is 1 (enabled).
 575
 576==============================================================
 577
 578oom_kill_allocating_task
 579
 580This enables or disables killing the OOM-triggering task in
 581out-of-memory situations.
 582
 583If this is set to zero, the OOM killer will scan through the entire
 584tasklist and select a task based on heuristics to kill.  This normally
 585selects a rogue memory-hogging task that frees up a large amount of
 586memory when killed.
 587
 588If this is set to non-zero, the OOM killer simply kills the task that
 589triggered the out-of-memory condition.  This avoids the expensive
 590tasklist scan.
 591
 592If panic_on_oom is selected, it takes precedence over whatever value
 593is used in oom_kill_allocating_task.
 594
 595The default value is 0.
 596
 597==============================================================
 598
 599overcommit_kbytes:
 600
 601When overcommit_memory is set to 2, the committed address space is not
 602permitted to exceed swap plus this amount of physical RAM. See below.
 603
 604Note: overcommit_kbytes is the counterpart of overcommit_ratio. Only one
 605of them may be specified at a time. Setting one disables the other (which
 606then appears as 0 when read).
 607
 608==============================================================
 609
 610overcommit_memory:
 611
 612This value contains a flag that enables memory overcommitment.
 613
 614When this flag is 0, the kernel attempts to estimate the amount
 615of free memory left when userspace requests more memory.
 616
 617When this flag is 1, the kernel pretends there is always enough
 618memory until it actually runs out.
 619
 620When this flag is 2, the kernel uses a "never overcommit"
 621policy that attempts to prevent any overcommit of memory.
 622Note that user_reserve_kbytes affects this policy.
 623
 624This feature can be very useful because there are a lot of
 625programs that malloc() huge amounts of memory "just-in-case"
 626and don't use much of it.
 627
 628The default value is 0.
 629
 630See Documentation/vm/overcommit-accounting and
 631security/commoncap.c::cap_vm_enough_memory() for more information.
 632
 633==============================================================
 634
 635overcommit_ratio:
 636
 637When overcommit_memory is set to 2, the committed address
 638space is not permitted to exceed swap plus this percentage
 639of physical RAM.  See above.
 640
 641==============================================================
 642
 643page-cluster
 644
 645page-cluster controls the number of pages up to which consecutive pages
 646are read in from swap in a single attempt. This is the swap counterpart
 647to page cache readahead.
 648The mentioned consecutivity is not in terms of virtual/physical addresses,
 649but consecutive on swap space - that means they were swapped out together.
 650
 651It is a logarithmic value - setting it to zero means "1 page", setting
 652it to 1 means "2 pages", setting it to 2 means "4 pages", etc.
 653Zero disables swap readahead completely.
 654
 655The default value is three (eight pages at a time).  There may be some
 656small benefits in tuning this to a different value if your workload is
 657swap-intensive.
 658
 659Lower values mean lower latencies for initial faults, but at the same time
 660extra faults and I/O delays for following faults if they would have been part of
 661that consecutive pages readahead would have brought in.
 662
 663=============================================================
 664
 665panic_on_oom
 666
 667This enables or disables panic on out-of-memory feature.
 668
 669If this is set to 0, the kernel will kill some rogue process,
 670called oom_killer.  Usually, oom_killer can kill rogue processes and
 671system will survive.
 672
 673If this is set to 1, the kernel panics when out-of-memory happens.
 674However, if a process limits using nodes by mempolicy/cpusets,
 675and those nodes become memory exhaustion status, one process
 676may be killed by oom-killer. No panic occurs in this case.
 677Because other nodes' memory may be free. This means system total status
 678may be not fatal yet.
 679
 680If this is set to 2, the kernel panics compulsorily even on the
 681above-mentioned. Even oom happens under memory cgroup, the whole
 682system panics.
 683
 684The default value is 0.
 6851 and 2 are for failover of clustering. Please select either
 686according to your policy of failover.
 687panic_on_oom=2+kdump gives you very strong tool to investigate
 688why oom happens. You can get snapshot.
 689
 690=============================================================
 691
 692percpu_pagelist_fraction
 693
 694This is the fraction of pages at most (high mark pcp->high) in each zone that
 695are allocated for each per cpu page list.  The min value for this is 8.  It
 696means that we don't allow more than 1/8th of pages in each zone to be
 697allocated in any single per_cpu_pagelist.  This entry only changes the value
 698of hot per cpu pagelists.  User can specify a number like 100 to allocate
 6991/100th of each zone to each per cpu page list.
 700
 701The batch value of each per cpu pagelist is also updated as a result.  It is
 702set to pcp->high/4.  The upper limit of batch is (PAGE_SHIFT * 8)
 703
 704The initial value is zero.  Kernel does not use this value at boot time to set
 705the high water marks for each per cpu page list.  If the user writes '0' to this
 706sysctl, it will revert to this default behavior.
 707
 708==============================================================
 709
 710stat_interval
 711
 712The time interval between which vm statistics are updated.  The default
 713is 1 second.
 714
 715==============================================================
 716
 717swappiness
 718
 719This control is used to define how aggressive the kernel will swap
 720memory pages.  Higher values will increase agressiveness, lower values
 721decrease the amount of swap.  A value of 0 instructs the kernel not to
 722initiate swap until the amount of free and file-backed pages is less
 723than the high water mark in a zone.
 724
 725The default value is 60.
 726
 727==============================================================
 728
 729- user_reserve_kbytes
 730
 731When overcommit_memory is set to 2, "never overommit" mode, reserve
 732min(3% of current process size, user_reserve_kbytes) of free memory.
 733This is intended to prevent a user from starting a single memory hogging
 734process, such that they cannot recover (kill the hog).
 735
 736user_reserve_kbytes defaults to min(3% of the current process size, 128MB).
 737
 738If this is reduced to zero, then the user will be allowed to allocate
 739all free memory with a single process, minus admin_reserve_kbytes.
 740Any subsequent attempts to execute a command will result in
 741"fork: Cannot allocate memory".
 742
 743Changing this takes effect whenever an application requests memory.
 744
 745==============================================================
 746
 747vfs_cache_pressure
 748------------------
 749
 750This percentage value controls the tendency of the kernel to reclaim
 751the memory which is used for caching of directory and inode objects.
 752
 753At the default value of vfs_cache_pressure=100 the kernel will attempt to
 754reclaim dentries and inodes at a "fair" rate with respect to pagecache and
 755swapcache reclaim.  Decreasing vfs_cache_pressure causes the kernel to prefer
 756to retain dentry and inode caches. When vfs_cache_pressure=0, the kernel will
 757never reclaim dentries and inodes due to memory pressure and this can easily
 758lead to out-of-memory conditions. Increasing vfs_cache_pressure beyond 100
 759causes the kernel to prefer to reclaim dentries and inodes.
 760
 761Increasing vfs_cache_pressure significantly beyond 100 may have negative
 762performance impact. Reclaim code needs to take various locks to find freeable
 763directory and inode objects. With vfs_cache_pressure=1000, it will look for
 764ten times more freeable objects than there are.
 765
 766==============================================================
 767
 768zone_reclaim_mode:
 769
 770Zone_reclaim_mode allows someone to set more or less aggressive approaches to
 771reclaim memory when a zone runs out of memory. If it is set to zero then no
 772zone reclaim occurs. Allocations will be satisfied from other zones / nodes
 773in the system.
 774
 775This is value ORed together of
 776
 7771       = Zone reclaim on
 7782       = Zone reclaim writes dirty pages out
 7794       = Zone reclaim swaps pages
 780
 781zone_reclaim_mode is disabled by default.  For file servers or workloads
 782that benefit from having their data cached, zone_reclaim_mode should be
 783left disabled as the caching effect is likely to be more important than
 784data locality.
 785
 786zone_reclaim may be enabled if it's known that the workload is partitioned
 787such that each partition fits within a NUMA node and that accessing remote
 788memory would cause a measurable performance reduction.  The page allocator
 789will then reclaim easily reusable pages (those page cache pages that are
 790currently not used) before allocating off node pages.
 791
 792Allowing zone reclaim to write out pages stops processes that are
 793writing large amounts of data from dirtying pages on other nodes. Zone
 794reclaim will write out dirty pages if a zone fills up and so effectively
 795throttle the process. This may decrease the performance of a single process
 796since it cannot use all of system memory to buffer the outgoing writes
 797anymore but it preserve the memory on other nodes so that the performance
 798of other processes running on other nodes will not be affected.
 799
 800Allowing regular swap effectively restricts allocations to the local
 801node unless explicitly overridden by memory policies or cpuset
 802configurations.
 803
 804============ End of Document =================================
 805
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.