ghg@pur-ee.UUCP (George Goble) (05/22/87)
We are Beta-testing a SCSI-bus cartridge tape made by a startup
company "Exabyte" in Boulder, CO. Now, only a handful of these drives
are in existance, however mass production is starting now.
This drive uses an 8MM "metal" tape VCR "Camcorder" cart
that stores 2.3 Gigabytes on a 2 hour tape. These VCR tapes are
available at your local Kmart or Target store. The VCR "deck" is
made in Japan, however, all the electronics is USA made. The heads
are slightly modified (3 heads) so read-after-write may be done.
No "video" is used in this system, Exabyte designed their own servo
and recording formats. Surface-mount construction is used, with
5 custom VLSI chips. No preformatting is needed.
This drive does full read-after-write and rewrites anything
with as much as a 1 bit error elsewhere on the tape (the controller
figures it all out). Exabyte has run thousands of passes
and found very little degradation in lab testing...I have run a
120 minute tape for about 1 month now and have had no unrecoverable
"hard" errors or badspots on the tape. Most of the problems have
been quirks in the SUN-2 SCSI interface and my driver bugs
(I understand the SUN-3 is much better)
Transfer rates are in the order of 15 Mbytes/min (drive streaming).
The "burst" transfer rate to the 256Kbyte internal buffer is around
1.5Mbytes/sec. A 120 minute tape will hold 2.3Gbytes and takes 2 hrs
45 mins to read/write (this is equiv to 16 reels of 2400' magtape
@ 6250 BPI). A near-EOT "warning" will be issued at around 2.0 Gbytes
(on a 120 min cart)..with 300 Mbytes of tape remaining (around 1/4"
on the supply spool). The transport moves the tape at around 0.5 in/sec
(slowed down 30% compared to a Camcorder deck since tracks can be
tolerated closer together than with video). An effective tape
speed of 150 in/sec is obtained due to the helical scan head
spinning at 1800 RPM. Linear recording density is 54,000 FCI giving
an aereal density of 44.2 M-FC per sq inch (819 tracks/inch).
(FCI is Flux-Changes-Inch)
The ECC is overkill. Each track or "stripe" contains eight 1-KByte
blocks of user data.. with 400 bytes of ECC appended to each block.
This ECC can correct a 263 byte burst error + around 80 1-2 bit
errors within a 1 Kbyte sector. Find another tape drive which does this!
There are no inter-record-gaps as in conventional magtape systems, Exabyte
looks more like a disk, with 1 Kbyte sectors. (size can be changed
with software though). However, there are Filemarks (tape marks, EOFs).
With current firmware being developed now, one will be able to
skip to a filemark at 10X read/write speed, however "record" or block
skips will still be at 1X read speed. A real pain if you are
using Berkeley dump/restore and need to get the last file off a
level 0 dump of a 1.3GB disk (1.3 hour wait to read up 1.3GB).
The drive is capable of moving the tape at 76X read speed
(rewind speed), but it cannot decode anything going by at that
speed. It takes 2 min 15 sec to transverse a 120 min tape in this mode.
Exabyte is thinking about implementing a "blind search"
or position mode which would allow one to issue a maint mode command to
start/stop the deck (at 76X read speed) and allow one to just
fastforward someplace. Dump tapes have the needed info so one
could figure out their corrent position by reading from the current spot.
It would be easy to modify restore to issue blind searches/reads to get to
the correct spot using a binary search in 3-4 mins worst case.
This is better than the search time on a 125 IPS 6250 BPI conventional
drive.
I am testing this drive at the moment on a SUN-2/120 and have a rough
cut of the driver sort of working.. (sundev/st.c and sc.c files were
modified). The drive uses +5 and +12V, is 5.25 inch formfactor size,
and includes the "controller" in the same package (ready to plug into
a SCSI bus).
In June, testing will move to a SUN-3, with the eventual goal
of doing filesystem backups via Ethernet (TCP-IP). A SUN-2 is
barely fast enough to generate/compare data fast enough to keep the
drive streaming. The drive is currently slightly faster (15Mbytes/min)
than network backups (~12 Mbytes/min on Gould-9080, CCI 6/32,
and other 5-10 MIP machines with fast disks and Enets)
Single quantity price will be around $3500, with discounts for quantity
and OEMs. Contact Exabyte at 303-442-4333 for more info. They are not
on USENET.
--ghg
George Goble
Electrical Engineering
Purdue University
W. Lafayette, Ind 47907
(317)-494-3545
uucp: {decvax,ucbvax,ihnp4}!pur-ee!ghg
arpa: ghg@purdue.edu