vm.txt 27 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749
  1. Documentation 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. For general info and legal blurb, please look in README.
  5. ==============================================================
  6. This file contains the documentation for the sysctl files in
  7. /proc/sys/vm and is valid for Linux kernel version 2.6.29.
  8. The files in this directory can be used to tune the operation
  9. of the virtual memory (VM) subsystem of the Linux kernel and
  10. the writeout of dirty data to disk.
  11. Default values and initialization routines for most of these
  12. files can be found in mm/swap.c.
  13. Currently, these files are in /proc/sys/vm:
  14. - admin_reserve_kbytes
  15. - block_dump
  16. - compact_memory
  17. - dirty_background_bytes
  18. - dirty_background_ratio
  19. - dirty_bytes
  20. - dirty_expire_centisecs
  21. - dirty_ratio
  22. - dirty_writeback_centisecs
  23. - drop_caches
  24. - extfrag_threshold
  25. - hugepages_treat_as_movable
  26. - hugetlb_shm_group
  27. - laptop_mode
  28. - legacy_va_layout
  29. - lowmem_reserve_ratio
  30. - max_map_count
  31. - memory_failure_early_kill
  32. - memory_failure_recovery
  33. - min_free_kbytes
  34. - min_slab_ratio
  35. - min_unmapped_ratio
  36. - mmap_min_addr
  37. - nr_hugepages
  38. - nr_overcommit_hugepages
  39. - nr_trim_pages (only if CONFIG_MMU=n)
  40. - numa_zonelist_order
  41. - oom_dump_tasks
  42. - oom_kill_allocating_task
  43. - overcommit_memory
  44. - overcommit_ratio
  45. - page-cluster
  46. - panic_on_oom
  47. - percpu_pagelist_fraction
  48. - stat_interval
  49. - swappiness
  50. - user_reserve_kbytes
  51. - vfs_cache_pressure
  52. - zone_reclaim_mode
  53. ==============================================================
  54. admin_reserve_kbytes
  55. The amount of free memory in the system that should be reserved for users
  56. with the capability cap_sys_admin.
  57. admin_reserve_kbytes defaults to min(3% of free pages, 8MB)
  58. That should provide enough for the admin to log in and kill a process,
  59. if necessary, under the default overcommit 'guess' mode.
  60. Systems running under overcommit 'never' should increase this to account
  61. for the full Virtual Memory Size of programs used to recover. Otherwise,
  62. root may not be able to log in to recover the system.
  63. How do you calculate a minimum useful reserve?
  64. sshd or login + bash (or some other shell) + top (or ps, kill, etc.)
  65. For overcommit 'guess', we can sum resident set sizes (RSS).
  66. On x86_64 this is about 8MB.
  67. For overcommit 'never', we can take the max of their virtual sizes (VSZ)
  68. and add the sum of their RSS.
  69. On x86_64 this is about 128MB.
  70. Changing this takes effect whenever an application requests memory.
  71. ==============================================================
  72. block_dump
  73. block_dump enables block I/O debugging when set to a nonzero value. More
  74. information on block I/O debugging is in Documentation/laptops/laptop-mode.txt.
  75. ==============================================================
  76. compact_memory
  77. Available only when CONFIG_COMPACTION is set. When 1 is written to the file,
  78. all zones are compacted such that free memory is available in contiguous
  79. blocks where possible. This can be important for example in the allocation of
  80. huge pages although processes will also directly compact memory as required.
  81. ==============================================================
  82. dirty_background_bytes
  83. Contains the amount of dirty memory at which the background kernel
  84. flusher threads will start writeback.
  85. Note: dirty_background_bytes is the counterpart of dirty_background_ratio. Only
  86. one of them may be specified at a time. When one sysctl is written it is
  87. immediately taken into account to evaluate the dirty memory limits and the
  88. other appears as 0 when read.
  89. ==============================================================
  90. dirty_background_ratio
  91. Contains, as a percentage of total system memory, the number of pages at which
  92. the background kernel flusher threads will start writing out dirty data.
  93. ==============================================================
  94. dirty_bytes
  95. Contains the amount of dirty memory at which a process generating disk writes
  96. will itself start writeback.
  97. Note: dirty_bytes is the counterpart of dirty_ratio. Only one of them may be
  98. specified at a time. When one sysctl is written it is immediately taken into
  99. account to evaluate the dirty memory limits and the other appears as 0 when
  100. read.
  101. Note: the minimum value allowed for dirty_bytes is two pages (in bytes); any
  102. value lower than this limit will be ignored and the old configuration will be
  103. retained.
  104. ==============================================================
  105. dirty_expire_centisecs
  106. This tunable is used to define when dirty data is old enough to be eligible
  107. for writeout by the kernel flusher threads. It is expressed in 100'ths
  108. of a second. Data which has been dirty in-memory for longer than this
  109. interval will be written out next time a flusher thread wakes up.
  110. ==============================================================
  111. dirty_ratio
  112. Contains, as a percentage of total system memory, the number of pages at which
  113. a process which is generating disk writes will itself start writing out dirty
  114. data.
  115. ==============================================================
  116. dirty_writeback_centisecs
  117. The kernel flusher threads will periodically wake up and write `old' data
  118. out to disk. This tunable expresses the interval between those wakeups, in
  119. 100'ths of a second.
  120. Setting this to zero disables periodic writeback altogether.
  121. ==============================================================
  122. drop_caches
  123. Writing to this will cause the kernel to drop clean caches, dentries and
  124. inodes from memory, causing that memory to become free.
  125. To free pagecache:
  126. echo 1 > /proc/sys/vm/drop_caches
  127. To free dentries and inodes:
  128. echo 2 > /proc/sys/vm/drop_caches
  129. To free pagecache, dentries and inodes:
  130. echo 3 > /proc/sys/vm/drop_caches
  131. As this is a non-destructive operation and dirty objects are not freeable, the
  132. user should run `sync' first.
  133. ==============================================================
  134. extfrag_threshold
  135. This parameter affects whether the kernel will compact memory or direct
  136. reclaim to satisfy a high-order allocation. /proc/extfrag_index shows what
  137. the fragmentation index for each order is in each zone in the system. Values
  138. tending towards 0 imply allocations would fail due to lack of memory,
  139. values towards 1000 imply failures are due to fragmentation and -1 implies
  140. that the allocation will succeed as long as watermarks are met.
  141. The kernel will not compact memory in a zone if the
  142. fragmentation index is <= extfrag_threshold. The default value is 500.
  143. ==============================================================
  144. hugepages_treat_as_movable
  145. This parameter is only useful when kernelcore= is specified at boot time to
  146. create ZONE_MOVABLE for pages that may be reclaimed or migrated. Huge pages
  147. are not movable so are not normally allocated from ZONE_MOVABLE. A non-zero
  148. value written to hugepages_treat_as_movable allows huge pages to be allocated
  149. from ZONE_MOVABLE.
  150. Once enabled, the ZONE_MOVABLE is treated as an area of memory the huge
  151. pages pool can easily grow or shrink within. Assuming that applications are
  152. not running that mlock() a lot of memory, it is likely the huge pages pool
  153. can grow to the size of ZONE_MOVABLE by repeatedly entering the desired value
  154. into nr_hugepages and triggering page reclaim.
  155. ==============================================================
  156. hugetlb_shm_group
  157. hugetlb_shm_group contains group id that is allowed to create SysV
  158. shared memory segment using hugetlb page.
  159. ==============================================================
  160. laptop_mode
  161. laptop_mode is a knob that controls "laptop mode". All the things that are
  162. controlled by this knob are discussed in Documentation/laptops/laptop-mode.txt.
  163. ==============================================================
  164. legacy_va_layout
  165. If non-zero, this sysctl disables the new 32-bit mmap layout - the kernel
  166. will use the legacy (2.4) layout for all processes.
  167. ==============================================================
  168. lowmem_reserve_ratio
  169. For some specialised workloads on highmem machines it is dangerous for
  170. the kernel to allow process memory to be allocated from the "lowmem"
  171. zone. This is because that memory could then be pinned via the mlock()
  172. system call, or by unavailability of swapspace.
  173. And on large highmem machines this lack of reclaimable lowmem memory
  174. can be fatal.
  175. So the Linux page allocator has a mechanism which prevents allocations
  176. which _could_ use highmem from using too much lowmem. This means that
  177. a certain amount of lowmem is defended from the possibility of being
  178. captured into pinned user memory.
  179. (The same argument applies to the old 16 megabyte ISA DMA region. This
  180. mechanism will also defend that region from allocations which could use
  181. highmem or lowmem).
  182. The `lowmem_reserve_ratio' tunable determines how aggressive the kernel is
  183. in defending these lower zones.
  184. If you have a machine which uses highmem or ISA DMA and your
  185. applications are using mlock(), or if you are running with no swap then
  186. you probably should change the lowmem_reserve_ratio setting.
  187. The lowmem_reserve_ratio is an array. You can see them by reading this file.
  188. -
  189. % cat /proc/sys/vm/lowmem_reserve_ratio
  190. 256 256 32
  191. -
  192. Note: # of this elements is one fewer than number of zones. Because the highest
  193. zone's value is not necessary for following calculation.
  194. But, these values are not used directly. The kernel calculates # of protection
  195. pages for each zones from them. These are shown as array of protection pages
  196. in /proc/zoneinfo like followings. (This is an example of x86-64 box).
  197. Each zone has an array of protection pages like this.
  198. -
  199. Node 0, zone DMA
  200. pages free 1355
  201. min 3
  202. low 3
  203. high 4
  204. :
  205. :
  206. numa_other 0
  207. protection: (0, 2004, 2004, 2004)
  208. ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  209. pagesets
  210. cpu: 0 pcp: 0
  211. :
  212. -
  213. These protections are added to score to judge whether this zone should be used
  214. for page allocation or should be reclaimed.
  215. In this example, if normal pages (index=2) are required to this DMA zone and
  216. watermark[WMARK_HIGH] is used for watermark, the kernel judges this zone should
  217. not be used because pages_free(1355) is smaller than watermark + protection[2]
  218. (4 + 2004 = 2008). If this protection value is 0, this zone would be used for
  219. normal page requirement. If requirement is DMA zone(index=0), protection[0]
  220. (=0) is used.
  221. zone[i]'s protection[j] is calculated by following expression.
  222. (i < j):
  223. zone[i]->protection[j]
  224. = (total sums of present_pages from zone[i+1] to zone[j] on the node)
  225. / lowmem_reserve_ratio[i];
  226. (i = j):
  227. (should not be protected. = 0;
  228. (i > j):
  229. (not necessary, but looks 0)
  230. The default values of lowmem_reserve_ratio[i] are
  231. 256 (if zone[i] means DMA or DMA32 zone)
  232. 32 (others).
  233. As above expression, they are reciprocal number of ratio.
  234. 256 means 1/256. # of protection pages becomes about "0.39%" of total present
  235. pages of higher zones on the node.
  236. If you would like to protect more pages, smaller values are effective.
  237. The minimum value is 1 (1/1 -> 100%).
  238. ==============================================================
  239. max_map_count:
  240. This file contains the maximum number of memory map areas a process
  241. may have. Memory map areas are used as a side-effect of calling
  242. malloc, directly by mmap and mprotect, and also when loading shared
  243. libraries.
  244. While most applications need less than a thousand maps, certain
  245. programs, particularly malloc debuggers, may consume lots of them,
  246. e.g., up to one or two maps per allocation.
  247. The default value is 65536.
  248. =============================================================
  249. memory_failure_early_kill:
  250. Control how to kill processes when uncorrected memory error (typically
  251. a 2bit error in a memory module) is detected in the background by hardware
  252. that cannot be handled by the kernel. In some cases (like the page
  253. still having a valid copy on disk) the kernel will handle the failure
  254. transparently without affecting any applications. But if there is
  255. no other uptodate copy of the data it will kill to prevent any data
  256. corruptions from propagating.
  257. 1: Kill all processes that have the corrupted and not reloadable page mapped
  258. as soon as the corruption is detected. Note this is not supported
  259. for a few types of pages, like kernel internally allocated data or
  260. the swap cache, but works for the majority of user pages.
  261. 0: Only unmap the corrupted page from all processes and only kill a process
  262. who tries to access it.
  263. The kill is done using a catchable SIGBUS with BUS_MCEERR_AO, so processes can
  264. handle this if they want to.
  265. This is only active on architectures/platforms with advanced machine
  266. check handling and depends on the hardware capabilities.
  267. Applications can override this setting individually with the PR_MCE_KILL prctl
  268. ==============================================================
  269. memory_failure_recovery
  270. Enable memory failure recovery (when supported by the platform)
  271. 1: Attempt recovery.
  272. 0: Always panic on a memory failure.
  273. ==============================================================
  274. min_free_kbytes:
  275. This is used to force the Linux VM to keep a minimum number
  276. of kilobytes free. The VM uses this number to compute a
  277. watermark[WMARK_MIN] value for each lowmem zone in the system.
  278. Each lowmem zone gets a number of reserved free pages based
  279. proportionally on its size.
  280. Some minimal amount of memory is needed to satisfy PF_MEMALLOC
  281. allocations; if you set this to lower than 1024KB, your system will
  282. become subtly broken, and prone to deadlock under high loads.
  283. Setting this too high will OOM your machine instantly.
  284. =============================================================
  285. min_slab_ratio:
  286. This is available only on NUMA kernels.
  287. A percentage of the total pages in each zone. On Zone reclaim
  288. (fallback from the local zone occurs) slabs will be reclaimed if more
  289. than this percentage of pages in a zone are reclaimable slab pages.
  290. This insures that the slab growth stays under control even in NUMA
  291. systems that rarely perform global reclaim.
  292. The default is 5 percent.
  293. Note that slab reclaim is triggered in a per zone / node fashion.
  294. The process of reclaiming slab memory is currently not node specific
  295. and may not be fast.
  296. =============================================================
  297. min_unmapped_ratio:
  298. This is available only on NUMA kernels.
  299. This is a percentage of the total pages in each zone. Zone reclaim will
  300. only occur if more than this percentage of pages are in a state that
  301. zone_reclaim_mode allows to be reclaimed.
  302. If zone_reclaim_mode has the value 4 OR'd, then the percentage is compared
  303. against all file-backed unmapped pages including swapcache pages and tmpfs
  304. files. Otherwise, only unmapped pages backed by normal files but not tmpfs
  305. files and similar are considered.
  306. The default is 1 percent.
  307. ==============================================================
  308. mmap_min_addr
  309. This file indicates the amount of address space which a user process will
  310. be restricted from mmapping. Since kernel null dereference bugs could
  311. accidentally operate based on the information in the first couple of pages
  312. of memory userspace processes should not be allowed to write to them. By
  313. default this value is set to 0 and no protections will be enforced by the
  314. security module. Setting this value to something like 64k will allow the
  315. vast majority of applications to work correctly and provide defense in depth
  316. against future potential kernel bugs.
  317. ==============================================================
  318. nr_hugepages
  319. Change the minimum size of the hugepage pool.
  320. See Documentation/vm/hugetlbpage.txt
  321. ==============================================================
  322. nr_overcommit_hugepages
  323. Change the maximum size of the hugepage pool. The maximum is
  324. nr_hugepages + nr_overcommit_hugepages.
  325. See Documentation/vm/hugetlbpage.txt
  326. ==============================================================
  327. nr_trim_pages
  328. This is available only on NOMMU kernels.
  329. This value adjusts the excess page trimming behaviour of power-of-2 aligned
  330. NOMMU mmap allocations.
  331. A value of 0 disables trimming of allocations entirely, while a value of 1
  332. trims excess pages aggressively. Any value >= 1 acts as the watermark where
  333. trimming of allocations is initiated.
  334. The default value is 1.
  335. See Documentation/nommu-mmap.txt for more information.
  336. ==============================================================
  337. numa_zonelist_order
  338. This sysctl is only for NUMA.
  339. 'where the memory is allocated from' is controlled by zonelists.
  340. (This documentation ignores ZONE_HIGHMEM/ZONE_DMA32 for simple explanation.
  341. you may be able to read ZONE_DMA as ZONE_DMA32...)
  342. In non-NUMA case, a zonelist for GFP_KERNEL is ordered as following.
  343. ZONE_NORMAL -> ZONE_DMA
  344. This means that a memory allocation request for GFP_KERNEL will
  345. get memory from ZONE_DMA only when ZONE_NORMAL is not available.
  346. In NUMA case, you can think of following 2 types of order.
  347. Assume 2 node NUMA and below is zonelist of Node(0)'s GFP_KERNEL
  348. (A) Node(0) ZONE_NORMAL -> Node(0) ZONE_DMA -> Node(1) ZONE_NORMAL
  349. (B) Node(0) ZONE_NORMAL -> Node(1) ZONE_NORMAL -> Node(0) ZONE_DMA.
  350. Type(A) offers the best locality for processes on Node(0), but ZONE_DMA
  351. will be used before ZONE_NORMAL exhaustion. This increases possibility of
  352. out-of-memory(OOM) of ZONE_DMA because ZONE_DMA is tend to be small.
  353. Type(B) cannot offer the best locality but is more robust against OOM of
  354. the DMA zone.
  355. Type(A) is called as "Node" order. Type (B) is "Zone" order.
  356. "Node order" orders the zonelists by node, then by zone within each node.
  357. Specify "[Nn]ode" for node order
  358. "Zone Order" orders the zonelists by zone type, then by node within each
  359. zone. Specify "[Zz]one" for zone order.
  360. Specify "[Dd]efault" to request automatic configuration. Autoconfiguration
  361. will select "node" order in following case.
  362. (1) if the DMA zone does not exist or
  363. (2) if the DMA zone comprises greater than 50% of the available memory or
  364. (3) if any node's DMA zone comprises greater than 70% of its local memory and
  365. the amount of local memory is big enough.
  366. Otherwise, "zone" order will be selected. Default order is recommended unless
  367. this is causing problems for your system/application.
  368. ==============================================================
  369. oom_dump_tasks
  370. Enables a system-wide task dump (excluding kernel threads) to be
  371. produced when the kernel performs an OOM-killing and includes such
  372. information as pid, uid, tgid, vm size, rss, nr_ptes, swapents,
  373. oom_score_adj score, and name. This is helpful to determine why the
  374. OOM killer was invoked, to identify the rogue task that caused it,
  375. and to determine why the OOM killer chose the task it did to kill.
  376. If this is set to zero, this information is suppressed. On very
  377. large systems with thousands of tasks it may not be feasible to dump
  378. the memory state information for each one. Such systems should not
  379. be forced to incur a performance penalty in OOM conditions when the
  380. information may not be desired.
  381. If this is set to non-zero, this information is shown whenever the
  382. OOM killer actually kills a memory-hogging task.
  383. The default value is 1 (enabled).
  384. ==============================================================
  385. oom_kill_allocating_task
  386. This enables or disables killing the OOM-triggering task in
  387. out-of-memory situations.
  388. If this is set to zero, the OOM killer will scan through the entire
  389. tasklist and select a task based on heuristics to kill. This normally
  390. selects a rogue memory-hogging task that frees up a large amount of
  391. memory when killed.
  392. If this is set to non-zero, the OOM killer simply kills the task that
  393. triggered the out-of-memory condition. This avoids the expensive
  394. tasklist scan.
  395. If panic_on_oom is selected, it takes precedence over whatever value
  396. is used in oom_kill_allocating_task.
  397. The default value is 0.
  398. ==============================================================
  399. overcommit_memory:
  400. This value contains a flag that enables memory overcommitment.
  401. When this flag is 0, the kernel attempts to estimate the amount
  402. of free memory left when userspace requests more memory.
  403. When this flag is 1, the kernel pretends there is always enough
  404. memory until it actually runs out.
  405. When this flag is 2, the kernel uses a "never overcommit"
  406. policy that attempts to prevent any overcommit of memory.
  407. Note that user_reserve_kbytes affects this policy.
  408. This feature can be very useful because there are a lot of
  409. programs that malloc() huge amounts of memory "just-in-case"
  410. and don't use much of it.
  411. The default value is 0.
  412. See Documentation/vm/overcommit-accounting and
  413. security/commoncap.c::cap_vm_enough_memory() for more information.
  414. ==============================================================
  415. overcommit_ratio:
  416. When overcommit_memory is set to 2, the committed address
  417. space is not permitted to exceed swap plus this percentage
  418. of physical RAM. See above.
  419. ==============================================================
  420. page-cluster
  421. page-cluster controls the number of pages up to which consecutive pages
  422. are read in from swap in a single attempt. This is the swap counterpart
  423. to page cache readahead.
  424. The mentioned consecutivity is not in terms of virtual/physical addresses,
  425. but consecutive on swap space - that means they were swapped out together.
  426. It is a logarithmic value - setting it to zero means "1 page", setting
  427. it to 1 means "2 pages", setting it to 2 means "4 pages", etc.
  428. Zero disables swap readahead completely.
  429. The default value is three (eight pages at a time). There may be some
  430. small benefits in tuning this to a different value if your workload is
  431. swap-intensive.
  432. Lower values mean lower latencies for initial faults, but at the same time
  433. extra faults and I/O delays for following faults if they would have been part of
  434. that consecutive pages readahead would have brought in.
  435. =============================================================
  436. panic_on_oom
  437. This enables or disables panic on out-of-memory feature.
  438. If this is set to 0, the kernel will kill some rogue process,
  439. called oom_killer. Usually, oom_killer can kill rogue processes and
  440. system will survive.
  441. If this is set to 1, the kernel panics when out-of-memory happens.
  442. However, if a process limits using nodes by mempolicy/cpusets,
  443. and those nodes become memory exhaustion status, one process
  444. may be killed by oom-killer. No panic occurs in this case.
  445. Because other nodes' memory may be free. This means system total status
  446. may be not fatal yet.
  447. If this is set to 2, the kernel panics compulsorily even on the
  448. above-mentioned. Even oom happens under memory cgroup, the whole
  449. system panics.
  450. The default value is 0.
  451. 1 and 2 are for failover of clustering. Please select either
  452. according to your policy of failover.
  453. panic_on_oom=2+kdump gives you very strong tool to investigate
  454. why oom happens. You can get snapshot.
  455. =============================================================
  456. percpu_pagelist_fraction
  457. This is the fraction of pages at most (high mark pcp->high) in each zone that
  458. are allocated for each per cpu page list. The min value for this is 8. It
  459. means that we don't allow more than 1/8th of pages in each zone to be
  460. allocated in any single per_cpu_pagelist. This entry only changes the value
  461. of hot per cpu pagelists. User can specify a number like 100 to allocate
  462. 1/100th of each zone to each per cpu page list.
  463. The batch value of each per cpu pagelist is also updated as a result. It is
  464. set to pcp->high/4. The upper limit of batch is (PAGE_SHIFT * 8)
  465. The initial value is zero. Kernel does not use this value at boot time to set
  466. the high water marks for each per cpu page list.
  467. ==============================================================
  468. stat_interval
  469. The time interval between which vm statistics are updated. The default
  470. is 1 second.
  471. ==============================================================
  472. swappiness
  473. This control is used to define how aggressive the kernel will swap
  474. memory pages. Higher values will increase agressiveness, lower values
  475. decrease the amount of swap.
  476. The default value is 60.
  477. ==============================================================
  478. - user_reserve_kbytes
  479. When overcommit_memory is set to 2, "never overommit" mode, reserve
  480. min(3% of current process size, user_reserve_kbytes) of free memory.
  481. This is intended to prevent a user from starting a single memory hogging
  482. process, such that they cannot recover (kill the hog).
  483. user_reserve_kbytes defaults to min(3% of the current process size, 128MB).
  484. If this is reduced to zero, then the user will be allowed to allocate
  485. all free memory with a single process, minus admin_reserve_kbytes.
  486. Any subsequent attempts to execute a command will result in
  487. "fork: Cannot allocate memory".
  488. Changing this takes effect whenever an application requests memory.
  489. ==============================================================
  490. vfs_cache_pressure
  491. ------------------
  492. Controls the tendency of the kernel to reclaim the memory which is used for
  493. caching of directory and inode objects.
  494. At the default value of vfs_cache_pressure=100 the kernel will attempt to
  495. reclaim dentries and inodes at a "fair" rate with respect to pagecache and
  496. swapcache reclaim. Decreasing vfs_cache_pressure causes the kernel to prefer
  497. to retain dentry and inode caches. When vfs_cache_pressure=0, the kernel will
  498. never reclaim dentries and inodes due to memory pressure and this can easily
  499. lead to out-of-memory conditions. Increasing vfs_cache_pressure beyond 100
  500. causes the kernel to prefer to reclaim dentries and inodes.
  501. ==============================================================
  502. zone_reclaim_mode:
  503. Zone_reclaim_mode allows someone to set more or less aggressive approaches to
  504. reclaim memory when a zone runs out of memory. If it is set to zero then no
  505. zone reclaim occurs. Allocations will be satisfied from other zones / nodes
  506. in the system.
  507. This is value ORed together of
  508. 1 = Zone reclaim on
  509. 2 = Zone reclaim writes dirty pages out
  510. 4 = Zone reclaim swaps pages
  511. zone_reclaim_mode is set during bootup to 1 if it is determined that pages
  512. from remote zones will cause a measurable performance reduction. The
  513. page allocator will then reclaim easily reusable pages (those page
  514. cache pages that are currently not used) before allocating off node pages.
  515. It may be beneficial to switch off zone reclaim if the system is
  516. used for a file server and all of memory should be used for caching files
  517. from disk. In that case the caching effect is more important than
  518. data locality.
  519. Allowing zone reclaim to write out pages stops processes that are
  520. writing large amounts of data from dirtying pages on other nodes. Zone
  521. reclaim will write out dirty pages if a zone fills up and so effectively
  522. throttle the process. This may decrease the performance of a single process
  523. since it cannot use all of system memory to buffer the outgoing writes
  524. anymore but it preserve the memory on other nodes so that the performance
  525. of other processes running on other nodes will not be affected.
  526. Allowing regular swap effectively restricts allocations to the local
  527. node unless explicitly overridden by memory policies or cpuset
  528. configurations.
  529. ============ End of Document =================================