norm@fluke.UUCP (Norm Seethoff) (09/11/86)
We are wrestling with a common magtape device naming scheme which can be
used across all of our UN*X systems, independent of vendor. It seems clear
that we need to retain the /dev/{mt mt0 mt4 mt8 mt12 rmt*} conventions
for compatibility with existing software. What we would like to do,
however, is create a parallel naming convention which would even make
sense to novice users.
We are presently considering:
Device Name
Function
Rewind On Close No Rewind On Close
1600 bpi buffered /dev/mt1600 /dev/nmt1600
1600 bpi raw /dev/rmt1600 /dev/nrmt1600
6250 bpi buffered /dev/mt6250 /dev/nmt6250
6250 bpi raw /dev/rmt6250 /dev/nrmt6250
The use of "n" and "r" obviously overloads the letters. One could
easily guess that "nrmt1600" was the nonrewinding buffered 1600 bpi
device, when in fact in the above scheme it also specifies raw I/O.
Can any of you with experience in this area offer suggestions?
We are very interested in what direction UN*X system vendors may be
taking, especially AT&T, SUN, and DEC.
Norm Seethoff
John Fluke Mfg. Co., Inc.
{decvax!microso | allegra | lbl-csam | uw-beaver}!fluke!normwesommer@mit-trillian.MIT.EDU (William Sommerfeld) (09/16/86)
In article <355@sputnik.fluke.UUCP> norm@fluke.UUCP (Norm Seethoff) writes: >We are wrestling with a common magtape device naming scheme which can be >used across all of our UN*X systems, independent of vendor. It seems clear >that we need to retain the /dev/{mt mt0 mt4 mt8 mt12 rmt*} conventions >for compatibility with existing software. What we would like to do, >however, is create a parallel naming convention which would even make >sense to novice users. > Rewind On Close No Rewind On Close > 1600 bpi buffered /dev/mt1600 /dev/nmt1600 > 1600 bpi raw /dev/rmt1600 /dev/nrmt1600 > 6250 bpi buffered /dev/mt6250 /dev/nmt6250 > 6250 bpi raw /dev/rmt6250 /dev/nrmt6250 > We have it set up *exactly* this way on our timesharing systems, and have found it to be quite convenient. - Bill
ed@mtxinu.UUCP (Ed Gould) (09/18/86)
In article <355@sputnik.fluke.UUCP> norm@fluke.UUCP (Norm Seethoff) writes: >We are wrestling with a common magtape device naming scheme which can be >used across all of our UN*X systems, independent of vendor. ... > >We are presently considering: > Device Name > Function > Rewind On Close No Rewind On Close > > 1600 bpi buffered /dev/mt1600 /dev/nmt1600 > 1600 bpi raw /dev/rmt1600 /dev/nrmt1600 > 6250 bpi buffered /dev/mt6250 /dev/nmt6250 > 6250 bpi raw /dev/rmt6250 /dev/nrmt6250 Be sure to allow for multiple tape drives. The scheme we use is /dev/[n][r]mt{0,1,2}.{800,1600,6250} -- Ed Gould mt Xinu, 2560 Ninth St., Berkeley, CA 94710 USA {ucbvax,decvax}!mtxinu!ed +1 415 644 0146 "A man of quality is not threatened by a woman of equality."
john@unisoft.UUCP (John Sovereign) (09/18/86)
In article <355@sputnik.fluke.UUCP> norm@fluke.UUCP (Norm Seethoff) writes: >We are wrestling with a common magtape device naming scheme.... I believe that beginning with SVR2.?, AT&T is pushing a new convention: subdirectories in /dev for the names of secondary storage devices, i.e., /dev/dsk, /dev/rdsk, /dev/mt, and /dev/rmt for blocked and non-blocked i/o on disks and magnetic tape, respectively. For disks, the individual nodes look something like: /dev/dsk/c0d0s0 controller 0, drive 0, slice 0 I don't remember the individual names given to the tape devices, but why not: /dev/rmt/n1600 no-rewind-on-close, 1600bpi John Sovereign lll-lcc!unisoft!john 415/644-1230 UniSoft Systems 739 Allston Way Berkeley, CA 94710
rer@hpfclj.HP.COM (Rob Robason) (09/19/86)
> It seems clear > that we need to retain the /dev/{mt mt0 mt4 mt8 mt12 rmt*} conventions > for compatibility with existing software. What we would like to do, > however, is create a parallel naming convention which would even make > sense to novice users. > The use of "n" and "r" obviously overloads the letters. One could > easily guess that "nrmt1600" was the nonrewinding buffered 1600 bpi > device, when in fact in the above scheme it also specifies raw I/O. In our next releases we will be adopting the System V naming conventions. These specifically address some of your concerns. In this approach there are sub-directories of /dev (/dev/mt for block devices, /dev/rmt for character) into which similarly named devices are installed. The naming convention itself is as mnemonic as that which you suggested, and is summarized as follows: # - sequentially assigned device number l|m|h - low(800), medium(1600) or high(6250) bpi [n] - optional flag for "no rewind on close" an example for a 1600 bpi character tape with no rewind is: /dev/rmt/0mn and the corresponding block device is: /dev/mt/0mn We are also supporting the use of these as defaults for tar and mt, falling back to the old names of the new ones are not found. Tar will try to use /dev/rmt/0m then fall back to /dev/rmt8, likewise, mt tries to use /dev/rmt/0mn, falling back to /dev/rmt12. If you are inclined to gag at this verbosity, think back to the time when YOU were a novice: How mnemonic did you find /dev/rmt8? Anyway, if you're inclined to avoid the NIH (not invented here) syndrome, this provides at least one solution with a precedent. Rob Robason, HP SSO (Systems Software Operation), Fort Collins (hplabs|ihnp4|hpfcse|csu-cs|hpbbn)!hpfcla!rer