🚀 go-pugleaf

In article <7348@batcomputer.tn.cornell.edu>, gallow@batcomputer.tn.cornell.edu (Steven Gallow) writes:
> 
>   A while back there was a discussion on 9 track tape drives for the
>  IBM RT.  I have called CFN Industries (Livermore California) for   
>  information on their drives.  I'm interested in getting quotes from
>  some other companies.  The drive has to be at least 1600/6250 BPI.
>  Does anyone have any information?
> 
>                                     Thanks in Advance
>                                     Steve Gallow

I am doing some contract work on a tape driver for the RT.
My client has the tape drive on the SCSI bus, which requires
the IBM SCSI adapter at ~$1100 and only supports differential
SCSI interfaces, which are somewhat rare.  I guess you could
use any old AT interface card if you wanted a direct path,
but we are using AIX and you would need your own VRM driver,
which is a pain to write and debug.  If you use the SCSI VRM
driver IBM supplies, you only need to fight with it for a
couple of weeks.  Of course, if you need to write odd length
records, you are out of luck, since VRM gags on most transfers
that are not multiples of four bytes.  And if you need to
do a SCSI operation other than READ or WRITE that moves more
than 6K of data, you get to write your own VRM driver, because
the IBM driver doesn't support it.

We are evaluating a Cipher M990 GCR CacheTape, with models
from HP and StorageTek to be tested soon.  The Cipher was
actually purchased from a Dickens Data near Atlanta, but
the only reason they can charge almost double the base
price of the drive is that they supplied a (sorta) working
AIX driver.  Sorta means that until the latest revision,
asking for say 32K bytes with 10K physical records would
give you 32K of data in your buffer, the driver merrrily
spanning records for you and hiding all details of blocking.
Limited ioctl() support for tape movement, too.

If anyone is starting to write a tape driver for AIX, note
that the manual for the SCSI VRM driver is wrong in the
bit positions for setting variable byte transfers: it
says bits 2 & 3, but it should be bits 1 & 2 (from a helpful
guy in Austin who was reading the source as we talked).
Amazing how well things worked after that!

If anyone is considering starting from scratch with SCSI
on the RT, they might look at the Adaptec 1542 host adapter.
It supports single-ended devices (which are much more common)
with burst mode from the host adapter to the RT memory of
up to 10MBytes/sec.  Very clean design with mailboxes.
And a Mr. Christer Bernerus at uunet!mcvax!utc.chalmers.se!bernerus
has written drivers for tape and disk under AOS4.3, so we
know the hardware works.  A VRM driver would be some extra
work, though.


-- 
A. Lester Buck		...!texbell!moray!siswat!buck