streaming tape device driver wanted (LONG)
Carl S. Gutekunst
csg at pyramid.UUCP
Tue Mar 25 18:23:51 AEST 1986
I'm posting a reply, since references to this drive come up periodically and
few people seem to understand the pitfalls.
In article <165 at ms.UUCP> mts at ms.UUCP (Martin Stanley) writes:
>I am attempting to write a device driver for a Cipher 525 streaming tape
>drive. This is the type of drive that looks like a (very large) floppy
>disk and attaches to an existing floppy disk controller....
>
>The main problem I am having is figuring out how to keep the thing
>streaming....
When I was at Cromemco, I worked closely with the developer who wrote their
Cipher 525 driver. Engineering management chose it out of thin air to be the
primary backup device. It seemed so perfect: standard SA400 interface (which
we had), reasonably fast, 21Meg per cartridge. And it was very inexpensive,
well under $500. Cost was *the* overriding concern. Unfortunately, getting the
drive to do anything useful required either dedicating the entire CPU (usually
impossible in a Unix box) or using a disk controller that cost 5 to 10 times
more than the drive.
- The 525 *always* streams; it's a very simple transport without any braking.
Thus "start/stop" means 300ms to slow the tape down, 300ms to reverse, time
to step back ahead of where you were, 300ms to slow down.... i.e. every time
you can't keep up with the drive you lose at least 1.2 seconds while it
repositions itself.
- The tape is physically organized into 254 20K tracks (called "streams") and
128 to 1024 byte sectors, like a disk; the tape moves when the drive is
"selected" and scans forward to the next "stream" when given a step pulse.
This proved to be an awkward way to address a tape, but the 525 is obviously
too slow at random seek to treat it like a disk.
- Timing is highly critical; keeping the tape rolling without taking 1.2 sec.
naps requires an essentially dedicated CPU. Characters must be supplied at
a steady 62K per second, except for short breathers at sector gaps; step
pulses must be issued at precisely the correct time. Interrupt service with
Unix was too slow; interleaving sectors didn't help. At Cromemco the main
68000 had to do all the work, so we used 300K buffers and kept interrupts
disabled for the entire transfer. (Yes, this confused the time-of-day clock
amongst other things.)
The ideal solution would be a disk controller based on a fast, programmable,
onboard CPU with large buffers and good DMA. Buffer fills could occur during
start/stop cycles, or double-buffering could be used to prevent the cycles
entirely. But such cards would increase the total cost to nearly that of a
"real" 1/4" drive plus controller.
- The drive seems to be unusually sensitive to tape tension; every time you
insert a new cartridge the drive runs the tape all the way forward and back
in a "tensioning cycle." This takes about 3 minutes.
- The 525 wears out tapes rapidly, much more quickly than the standard Cipher
or Archive cartridge drives I'm familiar with. It reminds me of the cheap
Okidata color printers, which cost only $250 or so to buy but go through
$10 ribbons like they were free.
- The tape eject mechanism is balky, and occasionally disgorged its inner
workings in a shower of flying pieces. The eject level is positioned such
that the tape painfully pinches your fingers on the way out.
- Assorted headaches: The bit rate is 500K bps; this is twice the normal for a
DSDD 5.25" drive. Some controllers can handle this; others cannot. The drive
has built in read-after-write, but it proved inadequate; a separate verify
pass was necessary. The 525 occasionally got confused and mindlessly rolled
the tape off the spools. (My friend became an expert at re-threading them.)
It draws considerably more power than a normal 5.25" disk drive; your power
supply may not be able to handle it.
In short, the Cipher 525 is intended strictly for *cheap* dedicated backup on
single-user systems. All things considered, given the low price and simplicity
of the interface, it still may be your best buy for backup. For me on my Unix
box, I'd rather fork over a few more dollars and buy a real 1/4" drive. (The
cost of tape wear may eventually make up the difference.)
--
A disclaimer, for once: The above comments reflect my personal experiences and
opinions, and do not reflect the policies or positions of either Cromemco,
Inc. or Pyramid Technology Corp. I've used many other Cipher products and been
satisfied with them.
--
Carl S. Gutekunst {allegra,cmcl2,decwrl,hplabs,topaz,ut-sally}!pyramid!csg
Pyramid Technology Corp, Mountain View, CA +1 415 965 7200
More information about the Comp.unix.wizards
mailing list