Disk Partitioning

How do I know what size to make my disk partitions?

This is one of the more often asked questions I hear. Usually the answer is "It depends", so here is my experience with partitioning Linux boxes for various applications over the last few years.

First of all it helps to know exactly what the file systems are all used for and where stuff goes. A good reference for this sort of thing is in the Linux Documentation Project's "System Administrator's Guide" or SAG. You can find a good bit of info on the file system here

Alternatively, if you have a copy of "A Practical Guide to Linux", then check out page 74.

Here is a brief rundown...

/ Root file system. Should just contain /bin, /sbin, /dev,
/root,
/lib, and /etc.
/usr Programmes and source code.
/var Variable data, such as spools, man pages, news and mail
queues, database data.
/boot Boot kernels.
/home User data and "stuff".
/tmp Temporary file locations

The / file system will never need to be more than 100Meg. Make it that.

The /usr file system will vary depending on how big your initial installation is and how much extra software you download. For a RedHat 6.2 minimal install you'll be needing about 250 to 300 Meg (typical server), and for a full install you need around 1.5 Gig (typical workstation). Other distributions will need more or less, but this is a good guide. Any extra software you download may also go in this file system, so if you are planning installing an office suite or a cad package, be aware it that it may go in here.

If you are installing software to build from a tar ball or installing software that isn't part of a vendor's distribution, like an RPM or a DEB is, you will probably want to install it in the /usr/local file system. This file system is usually left untouched by the installation or upgrade process of a linux distribution and is ideal for installing third party software. If you plan on doing a lot of this, a separate partition is a great idea, because if you want to do a re-install rather than an upgrade, you can simply tell your distribution not to format the /usr/local file system when installing and you will leave your third party software in tact. The format of the /usr/local file system is almost identical to the / file system. Handy huh?

/usr/local/bin and /usr/local/sbin are also the correct place to put any scripts you may write after you have your system up and running. This is preferable to placing them in /usr/bin and /usr/sbin or even /bin and /sbin, as these should really be static and left the way the distribution intended them. It also makes backing up a system much easier if all your locally created scripts are in one convenient place.

The /var file system is the most varying file system, hence its name. The function of the machine will determine how much you need. For a vanilla system, I recommend 400 Meg. This is usually sufficient for a workstation. If you are building a proxy server, you will need a separate partition, but preferably a separate disk, for /var/spool/squid. The same goes for a mail server, except the file systems of interest are /var/spool/mqueue and /var/spool/mail. The size of /var/spool/mail will depend on how much storage you want for user's mailboxes, and the size /var/spool/mqueue will depend on how much mail 'in transit' you wish to spool. Mail server's acting as a secondary MX might need a lot here.

There are other smaller directories in /var/spool that are of interest, so I would recommend a /var/spool of 300 to 500 Meg for any server application in conjunction with the /var of 400 Meg. For a workstation you may be able to use the 400Meg /var partition to house your /var/spool as well, but it may pay to enlarge it a bit.

/var/log, as the name suggests, is the final resting place for logs. Once again the size of this will depend on the function of the system, but as a general rule it is highly recommended that you have a separate /var/log to your /var partition, regardless of the machine's function. This way any stray system logs that fill up will have no effect on your system other than stopping logging. This goes for both servers and workstations alike. If you are running a heavily loaded proxy, mail or web server, you will need heaps and heaps of disk space here. Fully loaded proxy servers in peering arrangements can easily generate hundreds of thousands of bytes of log files an hour. The same goes for mail servers. The mail can come in and go out very quickly on a fast link, but the log files stay around. You also don't want a slash-dotted web site to fill up your logging directory, so careful thought here will pay off in the future.

The /var file system is also often used for the storage of database data. /var/db or /var/lib is the file system that is used, and you will need to keep this big enough to hold your data. Often a separate fast SCSI disk or RAID will make your database much faster. IO is often the biggest bottleneck in database systems, and an IDE drive in /var/db or /var/lib wont help.

The /boot directory is probably the most useful file system, and often the most forgotten. Having your kernels on a separate partition will make rescuing a system that has crashed a whole lot easier. This means that booting the system and recovering the partitions can be attacked as two separate tasks. Having a small /boot in a primary partition is also the best cure for the famous "I just installed linux and now all I get is 'LI'" LILO installation problems. LILO still has issues with hard drive space above 1024 cylinders. A small 20 Meg /boot partition as the first primary partition on the system will alleviate this. Some distributions, such as RedHat, are smart enough to assign automatically the first primary partition to /boot for just this reason.

/home is where you hang your hat. It is also where you "keep your stuff". Files you download, projects, mail, documents, mp3's, everything. This is the equivalent of Windows' "My Documents", "C:/download", the desktop, etc. Even if the system is only used by you at your desk, and no-one else, you should still have your own home directory in the /home file system. Don't be tempted to add partitions to the root file system such as /scripts, /downloads, etc. You are breaking stuff when you do that. Linux is still a true multiuser operating system, even if you are the only person using it. Try to keep this in mind when building a partition table. This all starts to make sense when you stop logging in as root, and start logging in as a regular user. It never ceases to amaze me how many people run X as root. *sigh*.

Many distributions nowadays are geared towards easy and quick upgrades and everything has it's place. If you keep you stuff in /home/yourname and no-where else, you can be sure that when your next upgrade of linux comes, you can just chuck in the CD and hit "upgrade" and your Metallica mp3's will still be there when your system comes back on-line.

/home is also where the storage file system for a file server should go. The same is true for web server pages, and ftp server data. Obviously if you are building a web server, have a separate /home/httpd file system on a nice fast SCSI disk. Same with /home/ftp.

Sometimes it's a great idea to have a separate /tmp directory, because temporary files can get out of control. Having /tmp on the same partition as the root file system can cause problems if you scan a 60 Meg picture into a graphics manipulation programme and it decides to store it in /tmp.

The only other partition of major interest is the swap partition. It is often a good idea to place this in the physical middle of the drive. Then the heads have less far to travel to swap out data when the system gets loaded. Alternatively you can just throw more memory at the problem.

Now I'll give you a few 'real life' examples of servers that I maintain. The names have been changed to protect the innocent.

Here is my bog standard workstation. It runs X. It may get used for some server functions in the future, so there is lot's of space ready. I even have a big block of space hanging of /mnt/tmp, and one day I'm sure I'll think of a use for it.

[alex@workstation alex]$ df
Filesystem 1k-blocks Used Available Use% Mounted on
/dev/hda13 85530 34264 46850 42% /
/dev/hda1 101089 6802 89068 7% /boot
/dev/hda6 1517920 154616 1286196 11% /home
/dev/hda12 2150420 20 2041160 0% /mnt/tmp
/dev/hda10 248895 27 236018 0% /tmp
/dev/hda5 2016016 1292380 621224 68% /usr
/dev/hda7 758936 37592 682792 5% /var
/dev/hda9 497829 657 471470 0% /var/log
/dev/hda8 758936 292 720092 0% /var/spool

This next beast is a mail server. Note the use of separate drives for critical server file systems.

[alex@mail alex]$ df
Filesystem 1k-blocks Used Available Use% Mounted on
/dev/sda12 79941 39339 36474 52% /
/dev/sda1 21011 5463 14463 27% /boot
/dev/sda11 701636 43332 622664 7% /home
/dev/sda9 202031 13 191587 0% /tmp
/dev/sda5 1210800 456856 692436 40% /usr
/dev/sda7 496695 7069 463981 2% /var
/dev/sda6 1009724 197880 760552 21% /var/log
/dev/sda8 496695 982 470068 0% /var/spool
/dev/sdb1 4382932 766640 3393648 18% /var/spool/mail

Here is a proxy server. Mix of SCSI and IDE.

[root@proxy /root]# df
Filesystem 1k-blocks Used Available Use% Mounted on
/dev/hda11 101485 28799 67446 30% /
/dev/hda1 23393 2647 19538 12% /boot
/dev/hda7 199085 2101 186704 1% /home
/dev/hda8 81954 985 76737 1% /tmp
/dev/hda5 809556 170444 597988 22% /usr
/dev/hda6 199085 4243 184562 2% /var
/dev/hda10 1611224 10408 1518968 1% /var/log
/dev/sda1 17654736 354220 16403692 2% /var/spool/squid

When partitioning a machine for use, it is often a bad idea to install everything into a single / partition. Even if you don't need separate partitions, the practice you get from partitioning disks and learning how much space each partition needs in a given situation will be invaluable when someone asks you to build a server for them. Spend a few minutes before installation considering the functions of the machine you are building and this will yield a useful and efficient partition table. The more often you do it the more of a feel you will get for how much space your distribution needs for different tasks. Now that you have the above information there is no excuse for poor partitioning, and you can help make the world a safer place for data!

Next time you see...

[lame@nothought /]$ df
Fileystem 1k-blocks Used Available Use% Mounted on
/dev/hda1 17654736 1354220 15403692 8% /

...you can do something about it!