1 Documentation for /proc/sys/vm/* kernel version 2.2.5
2 (c) 1998, 1999, Rik van Riel <riel@nl.linux.org>
4 For general info and legal blurb, please look in README.
6 ==============================================================
8 This file contains the documentation for the sysctl files in
9 /proc/sys/vm and is valid for Linux kernel version 2.2.
11 The files in this directory can be used to tune the operation
12 of the virtual memory (VM) subsystem of the Linux kernel, and
13 one of the files (bdflush) also has a little influence on disk
16 Default values and initialization routines for most of these
17 files can be found in mm/swap.c.
19 Currently, these files are in /proc/sys/vm:
29 ==============================================================
33 This file controls the operation of the bdflush kernel
34 daemon. The source code to this struct can be found in
35 linux/mm/buffer.c. It currently contains 9 integer values,
36 of which 6 are actually used by the kernel.
38 From linux/fs/buffer.c:
39 --------------------------------------------------------------
42 int nfract; /* Percentage of buffer cache dirty to
44 int ndirty; /* Maximum number of dirty blocks to
45 write out per wake-cycle */
46 int nrefill; /* Number of clean buffers to try to
47 obtain each time we call refill */
48 int nref_dirt; /* Dirty buffer threshold for activating
49 bdflush when trying to refill buffers. */
50 int dummy1; /* unused */
51 int age_buffer; /* Time for normal buffer to age before
53 int age_super; /* Time for superblock to age before we
55 int dummy2; /* unused */
56 int dummy3; /* unused */
58 unsigned int data[N_PARAM];
59 } bdf_prm = {{40, 500, 64, 256, 15, 30*HZ, 5*HZ, 1884, 2}};
60 --------------------------------------------------------------
62 The first parameter governs the maximum number of dirty
63 buffers in the buffer cache. Dirty means that the contents
64 of the buffer still have to be written to disk (as opposed
65 to a clean buffer, which can just be forgotten about).
66 Setting this to a high value means that Linux can delay disk
67 writes for a long time, but it also means that it will have
68 to do a lot of I/O at once when memory becomes short. A low
69 value will spread out disk I/O more evenly.
71 The second parameter (ndirty) gives the maximum number of
72 dirty buffers that bdflush can write to the disk in one time.
73 A high value will mean delayed, bursty I/O, while a small
74 value can lead to memory shortage when bdflush isn't woken
77 The third parameter (nrefill) is the number of buffers that
78 bdflush will add to the list of free buffers when
79 refill_freelist() is called. It is necessary to allocate free
80 buffers beforehand, since the buffers often are of a different
81 size than memory pages and some bookkeeping needs to be done
82 beforehand. The higher the number, the more memory will be
83 wasted and the less often refill_freelist() will need to run.
85 When refill_freelist() comes across more than nref_dirt dirty
86 buffers, it will wake up bdflush.
88 Finally, the age_buffer and age_super parameters govern the
89 maximum time Linux waits before writing out a dirty buffer
90 to disk. The value is expressed in jiffies (clockticks), the
91 number of jiffies per second is 100, except on Alpha machines
92 (1024). Age_buffer is the maximum age for data blocks, while
93 age_super is for filesystem metadata.
95 ==============================================================
98 The three values in this file correspond to the values in
99 the struct buffer_mem. It controls how much memory should
100 be used for buffer memory. The percentage is calculated
101 as a percentage of total system memory.
104 min_percent -- this is the minimum percentage of memory
105 that should be spent on buffer memory
106 borrow_percent -- UNUSED
107 max_percent -- UNUSED
109 ==============================================================
112 This file contains the values in the struct freepages. That
113 struct contains three members: min, low and high.
115 The meaning of the numbers is:
117 freepages.min When the number of free pages in the system
118 reaches this number, only the kernel can
119 allocate more memory.
120 freepages.low If the number of free pages gets below this
121 point, the kernel starts swapping agressively.
122 freepages.high The kernel tries to keep up to this amount of
123 memory free; if memory comes below this point,
124 the kernel gently starts swapping in the hopes
125 that it never has to do real agressive swapping.
127 ==============================================================
131 Kswapd is the kernel swapout daemon. That is, kswapd is that
132 piece of the kernel that frees memory when it gets fragmented
133 or full. Since every system is different, you'll probably want
134 some control over this piece of the system.
136 The numbers in this page correspond to the numbers in the
137 struct pager_daemon {tries_base, tries_min, swap_cluster
138 }; The tries_base and swap_cluster probably have the
139 largest influence on system performance.
141 tries_base The maximum number of pages kswapd tries to
142 free in one round is calculated from this
143 number. Usually this number will be divided
144 by 4 or 8 (see mm/vmscan.c), so it isn't as
146 When you need to increase the bandwidth to/from
147 swap, you'll want to increase this number.
148 tries_min This is the minimum number of times kswapd
149 tries to free a page each time it is called.
150 Basically it's just there to make sure that
151 kswapd frees some pages even when it's being
152 called with minimum priority.
153 swap_cluster This is the number of pages kswapd writes in
154 one turn. You want this large so that kswapd
155 does it's I/O in large chunks and the disk
156 doesn't have to seek often, but you don't want
157 it to be too large since that would flood the
160 ==============================================================
164 This value contains a flag that enables memory overcommitment.
165 When this flag is 0, the kernel checks before each malloc()
166 to see if there's enough memory left. If the flag is nonzero,
167 the system pretends there's always enough memory.
169 This feature can be very useful because there are a lot of
170 programs that malloc() huge amounts of memory "just-in-case"
171 and don't much of it.
173 Look at: mm/mmap.c::vm_enough_memory() for more information.
175 ==============================================================
179 The Linux VM subsystem avoids excessive disk seeks by reading
180 multiple pages on a page fault. The number of pages it reads
181 is dependent on the amount of memory in your machine.
183 The number of pages the kernel reads in at once is equal to
184 2 ^ page-cluster. Values above 2 ^ 5 don't make much sense
185 for swap because we only cluster swap data in 32-page groups.
187 ==============================================================
191 This file does exactly the same as buffermem, only this
192 file controls the struct page_cache, and thus controls
193 the amount of memory used for the page cache.
195 In 2.2, the page cache is used for 3 main purposes:
196 - caching read() data from files
197 - caching mmap()ed data and executable files
200 When your system is both deep in swap and high on cache,
201 it probably means that a lot of the swaped data is being
202 cached, making for more efficient swapping than possible
205 ==============================================================
209 The kernel keeps a number of page tables in a per-processor
210 cache (this helps a lot on SMP systems). The cache size for
211 each processor will be between the low and the high value.
213 On a low-memory, single CPU system you can safely set these
214 values to 0 so you don't waste the memory. On SMP systems it
215 is used so that the system can do fast pagetable allocations
216 without having to aquire the kernel memory lock.
218 For large systems, the settings are probably OK. For normal
219 systems they won't hurt a bit. For small systems (<16MB ram)
220 it might be advantageous to set both values to 0.