Re: Stuff from /bin, /sbin, /lib depending on /usr/lib libraries

[Jon Dowland <jmtd@debian.org>]

On Fri, Aug 31, 2012 at 10:57:10PM +0800, Thomas Goirand wrote:
> On 08/31/2012 06:55 PM, Riku Voipio wrote:
> > How is that different from having a botched / or /boot ? Why do you
> > think having a separate /usr will make / less prone to HD crashes?
> > You have / on RAID5 while /usr isn't?
> >   
> 
> Typically, I have / on 2 small RAID1 partitions making the array on the
> first
> 2 HDD (1 or 2 gigs), and /usr on a LVM on a much, much larger RAID array
> (I use mostly software RAID1 and RAID10, but in some cases, much bigger
> hardware RAID5). So yes, that's my usual server setup.
> 
> Also, / is a partition on which almost nothing is read or written, while
> the others (eg: /usr, /var, /tmp, swap) are a lot more I/O intensive.
> Which means that / is less prone to failure. Often, the 2nd RAID
> array gets degraded, but / isn't. So it does make a lot of sense to
> setup things this way, and yes, / is less prone to HD crashes this
> way (I'm talking from 10 years of experience running about 100
> servers this way, so it's not just theory, it's very practical experience).

I'm struggling to understand this. In the situation you outline (/ ok,
/usr, /var, /tmp, swap on another RAID which is hosed) -- whatever service
the machine was offering is surely not being offered anymore (/ being too
small to be useful for anything except a rescue environment).  So / surviving
whilst all your services/data are dead doesn't seem to be a big win to me at
all. Am I missing a detail?