Subject: Re: /rescue, crunchgen'ed?
To: Johnny Billquist <bqt@update.uu.se>
From: Brett Lymn <blymn@baesystems.com.au>
List: current-users
Date: 08/30/2002 23:47:24
On Fri, Aug 30, 2002 at 03:27:19PM +0200, Johnny Billquist wrote:
> 
> This whole discussion is absurd.
>

why prolong it then?
 
> 
> I'm actually amazed that people don't just accept this fact.
>

This could be because you are not seeing other people's point of
view.  You may not be entirely right.

> If we imagine
> the disk, I can point at any number of disk blocks which can become
> corrupt and I'll still be able to repair the system.
> (Admittedly there are *some* disk blocks that are vital, but not every
> disk block is).
> If you have a crunchgenned /rescue, any broken block will stop you, no
> matter which one.
> 

Yes that is easy to imagine... but to kill the system I would have to
imagine this broken block in concert with one that hits a critical
library or binary elsewhere on the disk.  We now have in our
imagination a disk with two separate areas of brokenness, that would
be a rather sick disk indeed.  If that was one of my machines I would
be reaching for a backup tape... pronto.  We can dream all sorts of
bad scenarios but the reality is that it is very unlikely that you
will get both your /rescue crunchgen and the other parts of the disk
nuked at the same time unless the disk is going down fast.

> How can people claim that it's just as robust is beyond me. The arguments
> borders on total silliness.
> 

Sorry... you are focussing on one small scenario, a rather unlikely
one at that.  You need to think of things like operator error, a bad
install, rogue programs that could damage the shared library system in
a manner that would render it unusable.  If this happens you can use
the tools in /rescue to try to put the pieces back together.  Your
focus on a hardware only failure is doing you a disservice.

> 
> I'll see if I can build my VAX systems totally static, if you really want
> to know, and see what performance gains I get.
>

You may do that if you wish but don't ask me to give up the advantages
that can be gained from a dynamically linked system... we can always
work towards improving the speed of the dynamic linking, if you just
keep everything static then you lose on flexibility (think full
relinks of all apps when a library changes), memory, disk space,
nsswitch, PAM,.... this list goes on - you lose a lot.

-- 
Brett Lymn