Best Filesystems?
Lennart Sorensen
lsorense-1wCw9BSqJbv44Nm34jS7GywD8/FfD2ys at public.gmane.org
Thu Aug 11 15:13:06 UTC 2005
On Wed, Aug 10, 2005 at 07:02:07PM -0400, CLIFFORD ILKAY wrote:
> I think putting all the OS and system files in one big partition is a
> poor choice when you have room to spare on your disk. I typically
> have the following partitions:
>
> /boot - 100M ext2, ro, nomount - no need for journalling if you don't
> mount the filesystem. The kernel is loaded before filesystems are
> mounted so whether this filesystem is mounted or not makes no
> difference.
>
> / - 300M ext3 - more than enough
>
> For the rest, I use logical volumes because it gives me the maximum
> flexibility. With conventional partitions, it is difficult to find
> the sweet spot for partition size. I always found myself either
> having too little or too much space. Using logical volumes gives me
> the ability to resize filesystems and volumes to find the optimal
> balance.
>
> swap - whatever size you need it to be. There is no hard and fast rule
> like in Windows where you have to make it X times the size of
> physical RAM. If you have loads of RAM, you do not need as much swap.
> If you do not have very much RAM, you may find it useful to have much
> more.
>
> /usr - size depends entirely on the machine - minimal installs, I
> allocate 300M, ext3. A development machine which has a full blown GUI
> and loads of developer tools might be 4GB. I can then mount /usr as
> ro and remount as rw if I need to install more software later.
Given how many packages install stuff to /usr that is wayyy to small.
> /usr/local - optional, mounted ro if present. If you have things that
> you are installing from tarball and want to preserve them if you do a
> reinstall/upgrade, you may want to have this as a separate partition.
>
> /var - 400M, ext3, rw
>
> /tmp - ext2 100M, rw on a personal machine, larger on a multiuser
> server - Who cares about journalling temp files? Note, this will be
> much too small for VMWare. VMWare's needs for temp file space grows
> over time as you run the virtual machine so it is difficult to
> estimate what it should be. VMWare Knowledgebase article 844 outlines
> some strategies for dealing with this. I installed VMWare in my home
> directory. I added: tmpDirectory = "/home/cilkay/vmware/tmp"
> to /etc/vmware/config. Since /home is huge, I never ran out of temp
> space on VMWare again.
Why not use tmpfs?
> /home - as big as you want it to be, ext3, rw.
>
> With LVM, there is little reason to soak up every bit of disk space
> right off the bat as you might do with conventional partitioning. You
> can leave unallocated space in the volume group and grow the volumes
> and filesystems as necessary down the road.
>
> My objective with all the partions is to isolate those things that
> change from those things that do not and to only mount those things
> which change as rw.
That sounds like one amazinly annoying setup to try and deal with when
upgrading anything. I would go nuts if I had to waste time remounting /
everytime I wanted to make a small config change and then have to
remount it ro afterwards. What a waste of time, for what I consider no
real gain.
I tend to do:
/boot (128M) (raid1/ext2)
/ 20G (raid1/ext3)
rest raid1 LVM
LVM containing /home /data and whatever else I think needs potentially
variable space. If I use something large like a web site or postgres or
something in /var, I move it to a dir on /data and symlink it to the
original place in /var.
Everything mounted rw so that I don't have to think about it when I want
to change something (I hate making a config change, just to have to
abort, change to rw, then go make the change again).
> ext3 seems to be a safe but lower performance choice. I've used
> ReiserFS and have not had any problems with it. Some swear by it
> while others swear at it. According to the Gentoo docs, XFS is
> inappropriate for machines which do not have fast disk arrays and are
> not connected to a UPS.
I used to swearh by reiserfs, then I swore at it when I realized (after
some data loss) how idiotic the design of it is. It HAS to fail in
certain not that uncommon situations. XFS just had too many memory
leaks and other bugs in 2.6 kernels lately, so I gave up on it. Also
ran much to slow with smaller files.
Lennart Sorensen
--
The Toronto Linux Users Group. Meetings: http://tlug.ss.org
TLUG requests: Linux topics, No HTML, wrap text below 80 columns
How to UNSUBSCRIBE: http://tlug.ss.org/subscribe.shtml
More information about the Legacy
mailing list