Re: Ideas for v2.1

Paul Slootman (paul@waubel.ahwau.ahold.nl)
Thu, 13 Jun 1996 18:52:12 +0000 (ZOM)


DUPRE Christophe <duprec@jsp.umontreal.ca> wrote:

> On Tue, 11 Jun 1996, Russell Steffen wrote:
>
> > Here's my wishlist for 2.1:
> >
> > 1) Reworking the disk subsystem, so that there is a single top-level
> > driver for all disks, with the lower level drivers (IDE and SCSI)
> > registering each valid disk (maybe each parition) with the top
> > level driver. With this scheme, we can go with a single major
> > device number for the disk subsystem with each partition with
> > it's own minor number. This means we can get rid of the /dev/hd*
> > and /dev/sd*, and replace them with a single set of devices (/dev/disk*?).

> Not really a good idea : Let's say you have IDE and SCSI disks:
> /dev/hda, /dev/hdb, /dev/sda and /dev/sdb
>
> With your system, you'd have:
> /dev/hda -> /dev/diska
> /dev/hdb -> /dev/diskb
> /dev/sda -> /dev/diskc
> /dev/sdb -> /dev/diskd
>
> Now one IDE disk is crashed and you remove it. Thus /dev/hdb no longer
> exists, and you get this mapping:
> /dev/hda -> /dev/diska
> /dev/sda -> /dev/diskb
> /dev/sdc -> /dev/diskc
>
> So now your /etc/fstab will mount the wrong partitions...

Actually, that happens now if one of your SCSI disks dies (and it's
not the one with the highest SCSI ID). /dev/sdc then becomes /dev/sdb.
I would much prefer the SCSI disks to be numbered in direct relation
to SCSI ID, such as /dev/sd13 (to indicate disk 3 on adapter 1, for
example). I'm prejudiced with this ever since I clobbered a file system
once because the (external) disk was called differently on system A than
it was on system B :-(

I assume the /dev/sdc in your example was a typo?

Paul Slootman

--
Murphy Software, Enschede, The Netherlands          There is no such thing
work: paul@wau.mis.ah.nl / paul@murphy.nl                as a coincidence.
home: paul@wurtel.hobby.nl