LDP/LDP/guide/docbook/sag/ch06.sgml

<chapter id="memory-management">
<title>Memory Management</title>

	<blockquote><para><quote>Minnet, jag har tappat mitt minne,
	&auml;r jag svensk eller finne, kommer inte ih&aring;g...</quote>
	(Bosse &Ouml;sterberg)
	</para>
	<para>A Swedish drinking song, (rough) translation: ``Memory, I
	have lost my memory.  Am I Swedish or Finnish?  I can't
	remember''</para>
	</blockquote>

	<para> This section describes the Linux memory management
	features, i.e., virtual memory and the disk buffer cache.
	The purpose and workings and the things the system administrator
	needs to take into consideration are described.</para>
	
<sect1 id="vm-intro">
<title>What is virtual memory?</title>

	<para>Linux supports <glossterm>virtual memory</glossterm>, that
	is, using a disk as an extension of RAM so that the effective
	size of usable memory grows correspondingly.  The kernel will
	write the contents of a currently unused block of memory to the
	hard disk so that the memory can be used for another purpose.
	When the original contents are needed again, they are read back
	into memory.  This is all made completely transparent to the
	user; programs running under Linux only see the larger amount of
	memory available and don't notice that parts of them reside on
	the disk from time to time.  Of course, reading and writing the
	hard disk is slower (on the order of a thousand times slower)
	than using real memory, so the programs don't run as fast.
	The part of the hard disk that is used as virtual memory is
	called the <glossterm>swap space</glossterm>.</para>

	<para>Linux can use either a normal file in the filesystem or a
	separate partition for swap space.  A swap partition is
	faster, but it is easier to change the size of a swap file
	(there's no need to repartition the whole hard disk, and
	possibly install everything from scratch).  When you know how
	much swap space you need, you should go for a swap partition,
	but if you are uncertain, you can use a swap file first, use
	the system for a while so that you can get a feel for how much
	swap you need, and then make a swap partition when you're
	confident about its size.</para>

	<para>You should also know that Linux allows one to use several swap
	partitions and/or swap files at the same time.  This means
	that if you only occasionally need an unusual amount of swap space,
	you can set up an extra swap file at such times, instead of
	keeping the whole amount allocated all the time.</para>
	
	<para>A note on operating system terminology: computer science 
	usually distinguishes between swapping (writing the whole process
	out to swap space) and paging (writing only fixed size parts,
	usually a few kilobytes, at a time). Paging is usually more
	efficient, and that's what Linux does, but traditional Linux
	terminology talks about swapping anyway.
	</para>

</sect1>

<sect1 id="swap-space">
<title>Creating a swap space</title>

	<para>A swap file is an ordinary file; it is in no way special
	to the kernel.	The only thing that matters to the kernel is that it
	has no holes, and that it is prepared for use with
	<command>mkswap</command>.  It must reside on a local disk, however;
	it can't reside in a filesystem that has been mounted
	over NFS due to implementation reasons.</para>

	<para>The bit about holes is important. The swap file reserves
	the disk space so that the kernel can quickly swap out a page
	without having to go through all the things that are necessary
	when allocating a disk sector to a file.  The kernel merely
	uses any sectors that have already been allocated to the file.
	Because a hole in a file means that there are no disk sectors
	allocated (for that place in the file), it is not good for the
	kernel to try to use them.</para>

	<para>One good way to create the swap file without holes is through
	the following command:

        <screen>
	<prompt>$</prompt> <userinput>dd if=/dev/zero of=/extra-swap bs=1024 
	count=1024</userinput>
	<computeroutput>1024+0 records in
	1024+0 records out</computeroutput>
	<prompt>$</prompt>
	</screen>

	where <filename>/extra-swap</filename> is the name of the swap
	file and the size of is given after the <literal>count=</literal>.
	It is best for the size to be a multiple of 4, because the
	kernel writes out <glossterm>memory pages</glossterm>, which
	are 4 kilobytes in size.  If the size is not a multiple of 4,
	the last couple of kilobytes may be unused.</para>

	<para>A swap partition is also not special in any way.	You create
	it just like any other partition; the only difference is that
	it is used as a raw partition, that is, it will not contain any
	filesystem at all.  It is a good idea to mark swap partitions
	as type 82 (Linux swap); this will the make partition listings
	clearer, even though it is not strictly necessary to the
	kernel.</para>

	<para>After you have created a swap file or a swap partition, you
	need to write a signature to its beginning; this contains some
	administrative information and is used by the kernel.  The
	command to do this is <command>mkswap</command>, used like this:

        <screen>
	<prompt>$</prompt> <userinput>mkswap /extra-swap 1024</userinput>
	<computeroutput>Setting up swapspace, size = 1044480 
	bytes</computeroutput>
	<prompt>$</prompt>
	</screen>

	Note that the swap space is still not in use yet: it exists,
	but the kernel does not use it to provide virtual memory.</para>
	
	<para>You should be very careful when using
	<command>mkswap</command>, since it does not check that the
	file or partition isn't used for anything else.  <emphasis>You
	can easily overwrite important files and partitions with
	<command>mkswap</command>!</emphasis> Fortunately, you should
	only need to use <command>mkswap</command> when you install
	your system.</para>

	<para>The Linux memory manager limits the size of each swap space to
	2 GB.  You can, however, use up to
	8 swap spaces simultaneously, for a total of 16GB.
	</para>

</sect1>

<sect1 id="using-swap">
<title>Using a swap space</title>

	<para>An initialized swap space is taken into use with
	<command>swapon</command>.  This command tells the kernel that
	the swap space can be used.  The path to the swap space is given
	as the argument, so to start swapping on a temporary swap file
	one might use the following command.

<screen>
<prompt>$</prompt> <userinput>swapon /extra-swap</userinput>
<prompt>$</prompt>
</screen>

	Swap spaces can be used automatically by listing them in
	the <filename>/etc/fstab</filename> file.

<screen>
/dev/hda8        none        swap        sw     0     0
/swapfile        none        swap        sw     0     0
</screen>

	The startup scripts will run the command <command>swapon
	-a</command>, which will start swapping on all the swap
	spaces listed in <command>/etc/fstab</command>.  Therefore,
	the <command>swapon</command> command is usually used only when
	extra swap is needed.</para>
	
	<para>You can monitor the use of swap spaces with
	<command>free</command>.  It will tell the total amount of swap
	space used.

<screen>
<prompt>$</prompt> <userinput>free</userinput>
<computeroutput>             total       used       free     shared   
 buffers
Mem:         15152      14896        256      12404       2528
-/+ buffers:            12368       2784
Swap:        32452       6684      25768</computeroutput>
<prompt>$</prompt>
</screen>

	The first line of output (<literal>Mem:</literal>) shows the
	physical memory.  The total column does not show the physical
	memory used by the kernel, which is usually about a megabyte.
	The used column shows the amount of memory used (the second
	line does not count buffers).  The free column shows completely
	unused memory.	The shared column shows the amount of memory
	shared by several processes; the more, the merrier.  The buffers
	column shows the current size of the disk buffer cache.</para>
	
	<para>That last line (<literal>Swap:</literal>) shows similar
	information for the swap spaces.  If this line is all zeroes,
	your swap space is not activated.</para>
	
	<para>The same information is available via
	<command>top</command>, or using the proc filesystem in file
	<filename>/proc/meminfo</filename>.  It is currently difficult
	to get information on the use of a specific swap space.</para>

	<para>A swap space can be removed from use with
	<command>swapoff</command>.  It is usually not necessary to do it,
	except for temporary swap spaces.  Any pages in use in the swap
	space are swapped in first; if there is not sufficient physical
	memory to hold them, they will then be swapped out (to some other
	swap space).  If there is not enough virtual memory to hold all
	of the pages Linux will start to thrash; after a long while it
	should recover, but meanwhile the system is unusable.  You should
	check (e.g., with <command>free</command>) that there is enough
	free memory before removing a swap space from use.</para>

	<para>All the swap spaces that are used automatically
	with <command>swapon -a</command> can be removed from use
	with <command>swapoff -a</command>; it looks at the file
	<filename>/etc/fstab</filename> to find what to remove.
	Any manually used swap spaces will remain in use.</para>

	<para>Sometimes a lot of swap space can be in use even though
	there is a lot of free physical memory.  This can happen for
	instance if at one point there is need to swap, but later a big
	process that occupied much of the physical memory terminates
	and frees the memory.  The swapped-out data is not automatically
	swapped in until it is needed, so the physical memory may remain
	free for a long time.  There is no need to worry about this,
	but it can be comforting to know what is happening.  </para>

</sect1>

<sect1 id="sharing-swap">
<title>Sharing swap spaces with other operating systems</title>

	<para>Virtual memory is built into many operating systems.
	Since they each need it only when they are running, i.e., never at
	the same time, the swap spaces of all but the currently running
	one are being wasted.  It would be more efficient for them to
	share a single swap space.  This is possible, but can require a
	bit of hacking.  The  Tips-HOWTO at
	<ulink url="http://www.tldp.org/HOWTO/Tips-HOWTO.html"> 
	http://www.tldp.org/HOWTO/Tips-HOWTO.html</ulink>, which contains 
	some advice on how to
	implement this.  </para>

</sect1>

<sect1 id="swap-allocation">
<title>Allocating swap space</title>

	<para>Some people will tell you that you should allocate twice as 
	much swap space as you have physical memory, but this is a bogus
	rule. Here's how to do it properly:

	<itemizedlist>

	<listitem>
	
	<para> Estimate your total memory needs.  This is the largest
	amount of memory you'll probably need at a time, that is the
	sum of the memory requirements of all the programs you want to
	run at the same time.  This can be done by running at the same
	time all the programs you are likely to ever be running at the
	same time.  </para>

	<para>For instance, if you want to run X, you should allocate
	about 8 MB for it, gcc wants several megabytes (some
	files need an unusually large amount, up to tens of
	megabytes, but usually about four should do), and so on.
	The kernel will use about a megabyte by itself, and the
	usual shells and other small utilities perhaps a few
	hundred kilobytes (say a megabyte together).  There is
	no need to try to be exact, rough estimates are fine,
	but you might want to be on the pessimistic side.</para>
	
	<para>Remember that if there are going to be several people
	using the system at the same time, they are all going
	to consume memory.  However, if two people run the same
	program at the same time, the total memory consumption
	is usually not double, since code pages and shared
	libraries exist only once.</para>
	
	<para>The <command>free</command> and <command>ps</command>
	commands are useful for estimating the memory needs.
	
	</listitem>

	<listitem>

	<para>Add some security to the estimate in step 1.  This is because
	estimates of program sizes will probably be wrong, because
	you'll probably forget some programs you want to run, and to
	make certain that you have some extra space just in case.  A
	couple of megabytes should be fine.  (It is better to allocate
	too much than too little swap space, but there's no need to
	over-do it and allocate the whole disk, since unused swap space
	is wasted space; see later about adding more swap.)  Also,
	since it is nicer to deal with even numbers, you can round the
	value up to the next full megabyte.</para>
	
	</listitem>
	
	<listitem>
	
	<para>Based on the computations above, you know how much memory
	you'll be needing in total.  So, in order to allocate swap
	space, you just need to subtract the size of your physical
	memory from the total memory needed, and you know how much
	swap space you need.  (On some versions of UNIX, you need to
	allocate space for an image of the physical memory as well, so
	the amount computed in step 2 is what you need and you shouldn't
	do the subtraction.)</para>

	</listitem>
	
	<listitem>

	<para>If your calculated swap space is very much larger than your
	physical memory (more than a couple times larger), you should
	probably invest in more physical memory, otherwise performance
	will be too low.</para>

	</itemizedlist>
    
	<para>It's a good idea to have at least some swap space, even if
	your calculations indicate that you need none. Linux uses
	swap space somewhat aggressively, so that as much physical
	memory as possible can be kept free. Linux will swap out
	memory pages that have not been used, even if the memory
	is not yet needed for anything. This avoids waiting for
	swapping when it is needed: the swapping can be done
	earlier, when the disk is otherwise idle.</para>

	<para>Swap space can be divided among several disks. This
	can sometimes improve performance, depending on the
	relative speeds of the disks and the access patterns
	of the disks. You might want to experiment with a few
	schemes, but be aware that doing the experiments
	properly is quite difficult. You should not believe
	claims that any one scheme is superior to any other,
	since it won't always be true.
	</para>
	
</sect1>

<sect1 id="buffer-cache">
<title>The buffer cache</title>

	<para>Reading from a disk
	is very slow compared to accessing (real) memory.  In addition,
	it is common to read the same part of a disk several times
	during relatively short periods of time.  For example, one
	might first read an e-mail message, then read the letter into
	an editor when replying to it, then make the mail program read
	it again when copying it to a folder.  Or, consider how often
	the command <command>ls</command> might be run on a system with
	many users.  By reading the information from disk only once
	and then keeping it in memory until no longer needed, one can
	speed up all but the first read.  This is called <glossterm>disk
	buffering</glossterm>, and the memory used for the purpose is
	called the <glossterm>buffer cache</glossterm>.</para>

	<para>Since memory is, unfortunately, a finite, nay, scarce
	resource, the buffer cache usually cannot be big enough (it
	can't hold all the data one ever wants to use).  When the cache
	fills up, the data that has been unused for the longest time
	is discarded and the memory thus freed is used for the new
	data.</para>

	<para>Disk buffering works for writes as well.	On the one hand,
	data that is written is often soon read again (e.g., a source
	code file is saved to a file, then read by the compiler),
	so putting data that is written in the cache is a good idea.
	On the other hand, by only putting the data into the cache, not
	writing it to disk at once, the program that writes runs quicker.
	The writes can then be done in the background, without slowing
	down the other programs.</para>

	<para>Most operating systems have buffer caches (although
	they might be called something else), but not all of
	them work according to the above principles.  Some are
	<glossterm>write-through</glossterm>: the data is written to disk
	at once (it is kept in the cache as well, of course).  The cache
	is called <glossterm>write-back</glossterm> if the writes are done
	at a later time.  Write-back is more efficient than write-through,
	but also a bit more prone to errors: if the machine crashes,
	or the power is cut at a bad moment, or the floppy is removed
	from the disk drive before the data in the cache waiting to be
	written gets written, the changes in the cache are usually lost.
	This might even mean that the filesystem (if there is one) is
	not in full working order, perhaps because the unwritten data
	held important changes to the bookkeeping information.</para>
	
	<para>Because of this, you should never turn off the
	power without using a proper shutdown procedure 
	or remove a floppy from the
	disk drive until it has been unmounted (if it was mounted)
	or after whatever program is using it has signaled that it
	is finished and the floppy drive light doesn't shine anymore.
	The <command>sync</command> command <glossterm>flushes</glossterm>
	the buffer, i.e., forces all unwritten data to be written to disk,
	and can be used when one wants to be sure that everything is
	safely written.  In traditional UNIX systems, there is a program
	called <command>update</command> running in the background
	which does a <command>sync</command> every 30 seconds, so
	it is usually not necessary to use <command>sync</command>.
	Linux has an additional daemon, <command>bdflush</command>,
	which does a more imperfect sync more frequently to avoid the
	sudden freeze due to heavy disk I/O that <command>sync</command>
	sometimes causes.</para>
	
	<para>Under Linux, <command>bdflush</command> is started by
	<command>update</command>.  There is usually no reason to worry
	about it, but if <command>bdflush</command> happens to die for
	some reason, the kernel will warn about this, and you should
	start it by hand (<command>/sbin/update</command>).</para>

	<para>The cache does not actually buffer files, but blocks, which
	are the smallest units of disk I/O (under Linux, they are usually
	1 KB).	This way, also directories, super blocks, other filesystem
	bookkeeping data, and non-filesystem disks are cached.</para>

	<para>The effectiveness of a cache is primarily decided by its
	size.  A small cache is next to useless: it will hold so little
	data that all cached data is flushed from the cache before it
	is reused.  The critical size depends on how much data is read
	and written, and how often the same data is accessed.  The only
	way to know is to experiment.</para>

	<para>If the cache is of a fixed size, it is not very good to have
	it too big, either, because that might make the free memory too
	small and cause swapping (which is also slow).	To make the most
	efficient use of real memory, Linux automatically uses all free
	RAM for buffer cache, but also automatically makes the cache
	smaller when programs need more memory.</para>

	<para>Under Linux, you do not need to do anything to make use
	of the cache, it happens completely automatically.  Except for
	following the proper procedures for shutdown and removing
	floppies, you do not need to worry about it.  </para>
</sect1>
</chapter>