home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!olivea!inews.Intel.COM!imutm1.de.intel.com!gold.sub.org!greenie!gert
- From: gert@greenie.gold.sub.org (Gert Doering)
- Newsgroups: comp.os.linux
- Subject: Re: SCSI Autodetect? Static table? (Was: ST01/02 and Syquest 5110)
- Message-ID: <1993Jan21.183902.12472@greenie.gold.sub.org>
- Date: 21 Jan 93 18:39:02 GMT
- References: <C0K0JH.5vx@mentor.cc.purdue.edu> <C0LGtv.AMD@ra.nrl.navy.mil> <1993Jan15.141951.9684@greenie.gold.sub.org> <C0x6JC.1Bw@ra.nrl.navy.mil>
- Organization: GreeniE
- Lines: 88
-
- eric@tantalus.nrl.navy.mil (Eric Youngdale) writes:
-
- >>I wrote:
-
- >>BTW, there is one detail I do not like at all in the numbering of the scsi
- >>hard disks.
- >>
- >>Why are the SCSI disks numbered
- >>
- >>/dev/sda* for the first disk (lowest SCSI ID)
- >>/dev/sdb* for the second disk (second lowest SCSI ID)
- >>/dev/sdc* ...
- >>
- >>and so on, regardless of the exact SCSI numbers? Imagine a system with
- >>three SCSI HDs, ID 0, 1 and 2, and HD 1 crashes. The Disk with ID 2,
- >>former /dev/sdc, is now /dev/sdb and the result is big confusion, mis-
- >>mounting, ...
- >>
- >>Why not naming them /dev/sd[i]* with "i" equal to the SCSI ID of the
- >>drive?
-
- > How would you treat tapes and cdroms? The same way? Then you be
- >forced to have /dev/st1 and /dev/scd2 if you had a /dev/sda0. Also, you can in
- >principle have more than one scsi controller in the system at the same time.
- >You are limited to one of each type, and there are probably some that conflict
- >with each other so you could have two or three different controllers. In fact
- >I think you could have one of nearly every type of scsi controller that linux
- >supports. How would you treat two disks with the same ID number that are on
- >different scsi controllers.
-
- Weeeeelll... right. I see your point.
-
- That's one thing I did not think about.
-
- But nevertheless, I think that confusing device names (-> they could always
- be linked to something more tasty ;-) ) are *better* than device names
- that change mysteriously just because you add or remove some *other*
- hardware in your system. I *know* about it - but what happens when a
- system administrator of a later Linux version (1.5?) who does *not* know
- it changes some HD and then erases the wrong drive? I think with the
- current naming scheme this can happen too fast.
-
- > Then again, how would you handle drive arrays where you access each
- >drive with a separate lun. Each drive would have the same ID number, but they
- >would all have different luns from each other.
-
- Ummm, good point.
-
- That's a point for SCO Unix - there you have a list of SCSI devices (host
- adapter number, SCSI ID and lun) and some bits of the device minors select
- drives out of this list.
-
- > The problem is not so much in the device naming, we could come up with
- >something that would be suitably confusing that would allow for all of the
- ^^^^^^^^^ :-)
- >possibilities. It comes up more in the major and minor numbering - there are
- >only 8 bits for the device minor number and we have to come up with a
- >consistent way of assigning all of the possible ID/lun combinations to the 256
- >available minor numbers (recall that each partition will get it's own minor
- >number). The only way to have fixed assignments would be to somehow restrict
- >the number of partitions per drive, the number of luns per id, the number of
- >disks per controller and/or the number of controllers on the system.
-
- Would there be a way to use different majors for the same driver? I think
- that's the way SCO is handling this. Nevertheless, if you want to be able
- to use *all* the possible combinations on the SCSI bus (7 devices, 8 luns
- each, 4+4 partitions each, multiple controllers) you end up with total
- chaos. What about limiting the number of devices on the (one) SCSI bus to
- 16 (4 bits), somehow divided (how?) between SCSI IDs and LUNs. Then you have
- 4 bits left to number partitions.
-
- For a second (third,...) SCSI host adaptor, you cold use a different major
- number (if that's possible).
-
- > The one possibile solution would be to allow you to fix the drive
- >designator (i.e. a, b, c, etc.) in the table that the scsi code reads at boot
- >time.
-
- This would solve the problem of the changing device names. But it would
- introduce others. Hmmm.
-
- How are the minor numbers used right now?
-
- gert
- --
- Gert Doering | SubNet : gert@greenie.gold.sub.org | mailbox / uucp:
- Munich / FRG | InterNet: gdoering@physik.tu-muenchen.de | call (089)3243328
- (089)3243228 | FidoNet : gert.doering@2:246/55.4 | login bbs / nuucp
-