[SGVLUG] Linux Partitioning for Server

Edgar Garrobo egarrobo at gmail.com
Tue Apr 7 17:51:27 PDT 2009


Thanks for the detailed feedback Rae.  I'm not so much worried about uptime
for this particular server.  The only way (in CentOS/Redhat) I can tell if
one of the 3Ware hardware RAID controllers notices a drive failure or RAID
problem is the loud alarm the card sounds.  Not the best way, but it
actually suffices for our little data center.

On Tue, Apr 7, 2009 at 4:13 PM, Rae Yip <rae.yip at gmail.com> wrote:

> There have been plenty of replies already, but I'll throw in my $0.02.
>
> I tend to agree with Tom; LVM is a wonderful thing. The main idea is
> to allow you to size your filesystems according to your actual needs,
> and worry about the physical layout separately.
>
> That said, if you're running this as a production server, I agree with
> others who have said that RAID 1 is a better way to go for the root
> disk; especially if you want to get decent performance out of your
> swap.
>
> You may think that getting the most capacity out of your drives is
> your priority, but with 4 disks, you're losing 1 disk to parity
> anyway, and paying a likely uptime penalty for not having a hot spare.
> So why not go with RAID1 and worry less? You'll only get 2N capacity
> instead of 3N, but 66% capacity on five 9's uptime is better than 100%
> on three 9's, if you care at all about that.
>
> As for sizing your filesystems, it's good practice to keep a separate
> /boot because you generally don't want to deal with fs corruption at
> the bootloader prompt. I tend not to give all space to / because you
> want a separate /var and to keep fsck times low, but 2gb is too small
> for modern distros.
>
> It's okay to keep /home part of / on a prod server, as long as your
> apps and data are stored on a separate fs (ie. /app or /srv); you may
> want to keep your app logs on yet another fs to keep them from
> crowding your data, but this depends on the app. So this looks
> something like:
>
> /boot      20-100MB
> /            20-40GB
> swap      1.5 - 2x RAM
> /var        6-15GB (more if you keep app logs here too)
> /app       rest of space - /app/log size
> /app/log  depends on app and desired retention
>
> That's only six logical partitions, not too much to remember. I would
> do things differently for a desktop (bigger separate /home, smaller
> /app).
>
> Finally I would ask, do you know how you can tell inside Linux that
> you've had a hardware RAID failure? I don't have much experience with
> PC RAID controllers, but if you can't tell when you need to replace a
> disk, then you have no business using hardware RAID. Factor in recent
> discussion about RAID5 not scaling to larger disks.
>
> -Rae.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.sgvlug.net/pipermail/sgvlug/attachments/20090407/0906b85f/attachment.html 


More information about the SGVLUG mailing list