[comp.periphs] Wanted: SCSI Disks for Sun3's

berger@datacube.UUCP (09/26/87)

Well  I  have  not  heard  anything  from  anyone  on  the  net about
connecting generic SCSI peripherals to Sun's.  

I want to add Disk and Tape to Sun 3 computers.   I would  like to do
it with as little pain as possible and the priceing should be in line
with the cost of the computer.   Currently  if I  get a  Sun 3/50 for
under  $5k,  I  have  to  pay  $8k  for 140Meg  Disk and  60 Meg tape
subsystem.  The  amount of  disk is  barely enough  and the subsystem
cost much more than the computer.  

The question is:  Can I hook generic SCSI  peripherals up  to the Sun
3's SCSI port?  I've been told by one  source that  we could purchase
Maxstor disks and either  Emulex or  Adaptec SCSI  controllers and it
should  all plug  and play  with the  standard sun  drivers.  Another
source though, said that there IS some magic in the firmware Sun uses
in  the SCSI  controlers in  their shoebox  disk subsystems.   If you
don't  have  the  special firmware,  you will  have some intermittant
errors.  Is this true?  

Another  possibility  would  be  to  take  one of  these lovely large
capacity winchester from Maxstor,  Rodine, Toshiba,  or Fujuitsu that
have built in  SCSI interfaces  and hook  them up  to the  Sun 3 SCSI
ports and have them work with the standard  Sun drivers?   This would
be a very cost effective solution.  

Has anyone successfully done any of the above?  Are there any tricks?

One last thing I'm looking for is an alternative source  for VME SCSI
interfaces  that  is  compatable  with  the  one  Sun  uses  in their
3/{140,110,160,260,280}  seires.    Or  a  company that  makes a high
performace SCSI VME interface that comes with drivers  that are known
to work with Suns.  

				Bob Berger 

Datacube Inc. Systems / Software Group	4 Dearborn Rd. Peabody, Ma 01960
VOICE:	617-535-6644;	FAX: (617) 535-5643;  TWX: (710) 347-0125
UUCP:	berger@datacube.COM,  ihnp4!datacube!berger
	{seismo,cbosgd,cuae2,mit-eddie}!mirror!datacube!berger

srg@quick.UUCP (09/28/87)

Here's the scoop.  Sun uses two types of disk controllers in their
shoeboxes for 3/50's and 3/60's.  The 71Mb disk is an ST506 type
with an Adaptec ACB4000A controller, and the 140Mb is an ESDI type
with a modified Emulex MD21 controller.  I think the mods to the
MD21 hardware are to make it look like an Adaptec, and Emulex is
apparently bound by contract not to deliver the modified firmware
to anybody but SUN, so count that out.  SUN apparently wrote their
driver around some Adaptec-specific extentions to SCSI, so you have
to use either one of their controller cards or one that emulates
such.  I tried a Western Digital WA1003A-SCS with no luck, although
WD swore it was completely compatable.  The Adaptec I got next has
worked flawlessly since I installed it, came with an informative
manual, and wasn't that much more (I paid $178+tax).  The adaptec
controller is capable of putting 18 sectors per track, but the
SUN kernel won't work properly with anything but 17.  Someone
slipped up somewhere and hard-coded the 17.  It almost works,
and I've told SUN about the problem, so maybe someday.  SUN
reserves SCSI id's 0-3 for disks (0 and 1 are gen'd in the
generic kernel) and uses 4 for the SCSI tape controller.  You
can put 2 drives per controller as units 0 and 1.  I'm using
a pair of Seagate ST4096's on one controller.  They work fine,
but get pretty warm (ie - you MUST have a fan).  If anyone finds
an attractive box with enough power and some baffling to keep
it quiet, PLEASE let me know about it.  I'm using Integrand
"little board" boxes.  They work, but sure aren't awe-inspiring.

newsuser@Valhall.UUCP (10/01/87)

In article <104400007@datacube> berger@datacube.UUCP writes:
>I want to add Disk and Tape to Sun 3 computers.   I would  like to do
>it with as little pain as possible and the priceing should be in line
>with the cost of the computer.   Currently  if I  get a  Sun 3/50 for
>under  $5k,  I  have  to  pay  $8k  for 140Meg  Disk and  60 Meg tape
>subsystem.  The  amount of  disk is  barely enough  and the subsystem
>cost much more than the computer.  

We have been using "generic" SCSI disks and streamers to 3/50's for
about 3 months now without any trouble. The disks are MICROPOLIS 1375
140 Mb formatted(2 per station) and ARCHIVE 5945S 1/4" streamer 60 Mb
(1 per station). Each disk/streamer is mounted in a box with
power supply and a fan.

There were no trouble at all with the installation of the hardware
(when we finally did get the latest version of the PROM for the disks :-)
) and the standard "diag" program was used to format the disks.

In Sweden its about 1/4th of the cost per unit using this
configuration instead of buying it from SUN!!

          
                     Richard Niklasson
_______________________________________________________________________________
Richard Niklasson             !  INTERNET:    rn@tts.lu.se
Dept of Communication Systems !  UUCP: ...!{uunet,mcvax,munnari}!enea!tts.lu!rn
Lund Institute of Technology  !  EARN/BITNET: erlangrn@seldc51
Box 118                       !  PHONE:   +46 46109008
S-221 00  LUND,  Sweden       !  TELEFAX: +46 46145823
-------------------------------------------------------------------------------

gnu@hoptoad.uucp (John Gilmore) (10/05/87)

Six months ago I heard from a (technically competent) friend that he'd
tried hooking up a CDC Wren 3 to his Sun-3/50 and it would not work.
However I have recently heard, from someone else, that they were
running four such drives on their 3/50.

When Sun started using SCSI, the only disk controller available that
had decent performance (e.g. 1:1 interleave) and was shipping was the
Adaptec ACB4000, so Sun's SCSI hardware and software were written and
debugged using it.  There was not much experience in the industry with
SCSI and this controller required some funny "manufacturer-specific"
commands to get it to work, e.g. to format the disk, and perhaps to
initialize it at power-up.  Later as Sun (and customers) tried hooking
up other things, they probably ended up generalizing the software; and
the newer SCSI controllers probably require few or no "manufacturer-
specific" commands.

Hooking up SCSI disks is the hardest case because you will want to boot
the system from them, which requires that the boot PROM chips on the
CPU board support your drive.  For most other devices, you can ignore
the device until the kernel is running, and it's easier to write a
kernel device driver than to change the PROM software.

I advise getting the absolute latest boot proms and latest system
release (3.4) from Sun if you want to hook up generic SCSI.  (If you
are on a service contract, I believe this is free.)  Even then, no
promises.  If you succeed, post something to Sun-Spots.
-- 
{dasys1,ncoast,well,sun,ihnp4}!hoptoad!gnu			  gnu@toad.com

jvs@micas.UUCP (Jo stockley) (10/09/87)

in article <128@quick.COM>, srg@quick.COM (Spencer Garrett) says:
> 
> 
> Here's the scoop.  Sun uses two types of disk controllers in their
> shoeboxes for 3/50's and 3/60's.  The 71Mb disk is an ST506 type
> with an Adaptec ACB4000A controller, and the 140Mb is an ESDI type
> with a modified Emulex MD21 controller.  I think the mods to the
> MD21 hardware are to make it look like an Adaptec, and Emulex is
> apparently bound by contract not to deliver the modified firmware
> to anybody but SUN, so count that out.

not so! We have been bying Emulex MD21 ESDI controllers from a supplier
and have been using them on the Sun 3/160 with both Micropollis and Siemens
discs (170 & 310 Mbyte) every since Sun first started supporting ESDI (which
for us was when we got release 3.2 of Sun UNIX). Adaptec produced an ESDI
SCSI interface that was supposed to emulate the ACB4000 but we were unable
to get it to work on the Sun without making changes to the sd.c driver
(which we do not have). Anybody who want's further info on any of this
contact me.

We also tried the ACB4070 rll ST506 interface before the ESDI support came
and it worked except that the Sun SCSI board is brain dammaged and cannot
handle the increased transfer rate (Also while I'm about it why the hell
did Sun ever produce a SCSI board that does nopt support DIS/RE CONNECT
and then supply it in a multi user system???

Oh well, thats my contribution, take it or leave it, I'm off to the pub!
Byeee
Jo.-- 
-------------
Jo. Stockley. (jvs@micas.uucp or ...!mcvax!ukc!micas!jvs)

Nodecrest Computer Systems Ltd
Byfleet, UK.

Phone: +44 09323 40555

chapman@eris.BERKELEY.EDU (Brent Chapman) (10/12/87)

In article <536@micas.UUCP> jvs@micas.UUCP (Jo stockley) writes:
>Also while I'm about it why the hell
>did Sun ever produce a SCSI board that does nopt support DIS/RE CONNECT
>and then supply it in a multi user system???

I seem to remember reading in the SunOS 3.4 Release Manual that this feature
is added in that release.  I don't worry much about my SCSI peripherals, so
I don't know if what you're talking about is the same as what is mentioned
in the 3.4 Release Manual.


-Brent
--
Brent Chapman				Senior Programmer/Analyst
koala!brent@lll-tis.arpa		Capital Market Technology, Inc.
lll-tis!koala!brent			1995 University Ave., Suite 390
Phone: 415/540-6400			Berkeley, CA  94704

mangler@cit-vax.Caltech.Edu (Don Speck) (10/15/87)

In article <536@micas.UUCP> jvs@micas.UUCP (Jo stockley) writes:
>Also while I'm about it why the hell
>did Sun ever produce a SCSI board that does nopt support DIS/RE CONNECT
>and then supply it in a multi user system???

I recently had a chance to benchmark SCSI I/O on a Sun 3/60FC
running SunOS 3.4 with Sun 141 MB ESDI disks.

Running 'dump' to /dev/null (most disk-intensive program handy)
on one disk at a time did 34 transfers per second; two at once
did 30 transfers per second per disk:  evidence of properly
working disconnect/reconnect.  dump or tar (!) to SCSI tape
would stream using the default (10K) blocksize, but not using
the Sun-recommended 63K blocksize, implying that the (Emulex)
SCSI tape controller has a small write-behind buffer.  4.3bsd
dump (overlaps disk & tape I/O) would stream the tape with any
blocksize, further evidence of disconnect/reconnect working.

I measured the minimum time between 1K reads and larger reads,
from which I derive the value of 'rotdelay' (about 8ms) and,
more importantly, break down the blocksize-dependent part of
read setup and recovery time.  Recovery time had a significant
dependence on blocksize; assuming this is due to DMA having
to catch up with the disk transfer rate, it implies a DMA
rate of 0.8 MB/s.  Disk controller buffering was sufficient
to not 'blow a rev' in the middle of a transfer, even on reads
of a couple of tracks.

Conclusion:  though not as fast as an Eagle, it's comparable
in speed to typical big disks such as the DEC RA81.

Don Speck   speck@vlsi.caltech.edu  amdahl!cit-vax!speck