Subject: Re: Separate /usr, etc...
To: None <>
From: Chuck Yerkes <>
List: current-users
Date: 12/16/2002 22:48:27
Well, there's production machines, and build/development machines.

I screw up production machines and it might mean a plane ride
(or at least some lag while I attempt a rebuild via a serial
port and a diskless boot).

Quoting Perry E. Metzger (
> Chuck Yerkes <> writes:
> > Why a separate root in general?
> > 1) "Whoops, was I in /usr/lib/ when I typed 'rm l*'?"
> >    More realistically, it meant that the junior system admin
> >    (or tired sr) has to take an action before making /usr/ rw.
> >    Or that hack to the FooDaemon that lets me overwrite /usr/bin/vi.
> I have never done such a thing, and besides, these days, replacing
> /usr/lib is a couple of minutes work.

Did it once, but I've also recovered machines where people have
aimed well at the foot and succeeded.  System admins don't always
have the control they want of developers' machines, but we do
have the responsibility to keep them up.

> > 2) /var/ MUST be read write (recall: we're not talking cf appliance
> >    machines).  Nothing like having logs or lots of mail take up
> >    that last disk space on root.
> I tend to make everything the one partition, so that is pretty easy. :)

Cool, so I cause you to log like hell, your monopartition fills
up.  I know most Unix apps react really well to running out of space.

> >    Well, in systems used by groups of people (ie. not at home), /usr
> > and /usr/local (oft the same) grow and this gets replaced and someone
> > needs that and put it in and, oh where did the space go?
> Thats why I like having one big partition. On a 120G drive, I don't
> like deciding /usr/pkg needs N gigs and /home needs M gigs etc. I just
> make it one big partition and I rarely have trouble. I used to spend
> lots of time symlinking crud back and forth because of space
> misallocations -- it no longer happens...

Realistically, on a machine here, I have 
root         50MB
/var        150MB
/usr         1 GB
/home       10 GB  (rest of disk).
misc other stuff for chroot web and different data

source and pkgsrc is under /home/.  Why?  Cause thats the
partition that's read/write and contains most of the change.
/usr hasn't been written to for several months.

Machines far away from me don't get or want source.  Hell, it
doesn't need static libraries or .h files :)

> > 40GB root scare me and lead people to really sloppy system admin.
> I blow away my machines rather frequently -- it is unusual for
> something to remain lying about for long.

Different roles.  At my last job, which Perry tried to talk me out
of :), we had boxes with mono-partitions.  Took them forever (weeks)
to figure out that some program was running away and filling up
gigabytes of useless logs.  With a /var of a size they expected it
to need, they would have quickly noted that, "hmmm, this has filled
its partition really quickly, something is amiss."

Management and digesting of logs and keeping machines under control
is both an art and luck.  In the real world of 3 people doing what 7
used to do, system admins often don't get to know the details of
the apps they are installing.  I've had news servers  turned off
cause they couldn't get it to behave and didn't have time to really
fix it.  CVS trumped NNTP on the "make it not suck" scale.

100GB partitions don't lead to good disk discipline.

How to start?

I have a new Suse Linux disk set here.  I have no idea what it
wants.  So I'll install it really quickly - just a rough guess of
what I'd need, run it for a day, get a notion of disk partition
sizes and then reinstall it in a couple days.

With 18GB disks being the smallest you can get, I have clients
who run slightly generous sized partitions, but some of these
have gone though several upgrades (RH 6.3 -> 7.1 -> 8.0) without

Bottom line:  To each his own, but my way is best ;-)