From dew Tue Nov 20 14:21 MST 1990
Received: by hpfclw.HP.COM; Tue, 20 Nov 90 14:20:31 mst
Date: Tue, 20 Nov 90 14:20:31 mst
From: Dave Willis <dew>
Full-Name: Dave Willis
To: dan@hpfcmb, harris@hpfcrlh
Subject: MACE TEAC floppy on PWS
Cc: bayes@hpisla, chris@hpfcrjn, cindy_furst@hp4020, david_palermo@hp0900,
        dew@hpfclw, gieso@hpisla, greer@hpisla, jab@hpisla, jwh@hpfclw,
        jws@hpfclw, kinsell@hpfcmb, mlc@hpfcrt, mlk@hpfcoll, quist@hpfclw,
        schink@hpisla, sjo@hpsjo, thb@hpfcma
Status: RO


Results of a Quick Test of the MACE TEAC SCSI floppy on Pascal Workstations
11/20/90
David P. Willis


MANAGEMENT SUMMARY
=================

The TEAC SCSI floppy driver currently being considered for the MACE
product does not automatically sense the media type of the floppy disc
inserted (which is a basic concept of the SCSI philosophy), and as such
will force unacceptable modifications to the operating systems that are
the primary targets of this platform.  As other TEAC SCSI floppy models
are available that perform this basic function, I find the current
choice unacceptable.

Both Rocky Mountain Basic (RMB) and Pascal Workstation (PWS) have an
explicit assumption that floppy drives will automatically sense the
media-type, i.e.  the block size, #blocks/track, and #tracks, and report
this information to the operating system.  All CS80/SS80 floppies have
provided this capability.  The TEAC SCSI floppy under investigation to
be used with the MACE hardware, model #FD-235HF-700-U, does not provide
this basic capability.  Such capability is a fundamental concept of the
SCSI philosophy.  The impact to these very stable and minimally staffed
operating systems is very high, and should be considered, especially in
lieu of the fact that other TEAC SCSI floppy models, namely FD-235HS-300
and FD-235JS-401, do automatically sense the media type and have been
shown to work on the Pascal Workstation w/out change, and, furthermore,
because MACE's primary target audience are RMB and PWS users.


SIDE NOTE
=========

The full impact to HP-UX is as yet undetermined.  However, I can not see
how the impact could be anything less than significant.



INVESTIGATION REPORT
====================

MACE SCSI H/W
=============

Dan Swanson provided the PWS team with a special TEAC SCSI floppy for
testing on PWS.  PWS was chosen because its SCSI driver is the as that
used in both RMB and the BOOT ROM, and SCSI expertise was available
there.

Two questions were considered:

	1) Can the Pascal Workstation talk to LIF discs formatted under
	a CS80 driver with the TEAC SCSI floppy?

	2) Can PWS boot from a set of such discs, using the TEAC SCSI
	floppy?

If the answer to question 1 is FALSE, then the answer to question number
2 is also FALSE.

In attempting to determine that answer to question #1, PWS was brought
up on a known bootable system that has the SCSI drivers installed.  The
MACE TEAC SCSI floppy was attached to the system, but a disc was not
inserted.  PWS recognized the presence of the TEAC floppy and configured
itself correctly, i.e.  it was correctly attached to UNIT #3.

Pre-formatted media with the LIF directory structure, and known data
files, was placed in the TEAC SCSI floppy and a FILER VOLUMES command
was executed.  This failed.  The reason for the failure was that the
pre-formatted media had 256 byte blocks, yet the TEAC SCSI floppy
reported to the system a device with 512 byte blocks.  When a read
attempt was made, the drive checksum and the checksum laid down at the
end of each block did not match (understandably so given the situation),
and the drive reported a BLOCK PARITY ERROR to PWS.

Both PWS and RMB expect that the physical device can successfully report
the size of the floppy disc inserted into the drive.  All previous
CS80/SS80 floppies have provided this capability.  The Magneto-Optical
SCSI drives also support this capability.  There currently does not
exist in either operating system a provision for handling floppy drives
that do not adhere to this basic assumption.

In my opinion, the behaviour exhibited by this TEAC SCSI floppy drive
violates the entire spirit of the SCSI philosophy, if not the letter of
the SCSI specification.  SCSI was designed to push the burden of
managing physical blocks away from the operating system, and to place
that burden on the physical disc driver.  In fact, the operating system
only deals in LOGICAL BLOCK SIZES when speaking to all types of SCSI
devices.


OTHER TEAC FLOPPY ALTERNATIVES
==============================

At our sister division in Queensferry, engineers have tested various
TEAC SCSI floppies on PWS.  They have chosen a TEAC floppy model for
their application which does not display the undesirable behaviour
demonstrated by the TEAC floppy under investigation for the MACE H/W.

Following is a letter received from Andy Batham of QTD R&D regarding
their investigation into TEAC floppies:

	The TEAC floppy we are using is the TEAC FD-235HS-300 which
	includes the TEAC FC-1-00 'FDD SCSI Interface Unit', i.e.
	control board.  I've had no trouble with this device.  By
	setting hardware straps it can be made to automatically sense 1M
	byte and 2Mbyte discs or can be hard strapped to one or other.

	I am also using the TEAC FD-235JS-401 which is a 4M byte floppy.
	Again it works fine.  We won't be using this in a product
	however since we have to be backwards compatible with existing
	systems.

	Apollo division (contact Bill Clemmey) are also investigating a
	TEAC SCSI floppy but model FD-235HS-302 which includes the
	FC-1-01 SCSI control board.  This control board does not permit
	the automatic media-sense function we need, it has to be set to
	the media type by a Mode Select command.

	One additional point:  TEAC are replacing both versions of the
	FD-235-HS in the new year with drop in replacements.  The new
	spec is better in places (shock is certainly).  We are awaiting
	samples.




TEAC SCSI FLOPPY SUPPORT IN PWS & RMB
=====================================

As MACE is primarily targeted to RMB and PWS customers, the impact to
these operating systems should be considered before making a decision on
the final MACE configuration.  As both of these operating systems
currently have minimal staffing and support, making a decision that will
force large scale modifications to the operating system can only damage
sales and supportability of MACE.

In the case of PWS (which can be mirrored in both RMB and the boot ROM),
supporting this TEAC SCSI floppy would mean modifying the SCSI TM such
that whenever a new disc is paced in the drive, the disc drive would be
told the disc has block size x and capacity y (where x and y would come
from a predetermined set), and then a read would be attempted.  This
would continue until the x,y set is exhausted or a read is successful.
Unfortunately, as the SCSI driver timeout value is of necessity very
large, on the order of seconds, this algorithm could take on the order
of minutes every time a disc is changed.

As it is not possible to tell the difference betwen Magneto-Optical
drives and other floppy drives, existing customers using the SCSI
Magneto-Optical drives would see a performance degradation due to this
algorithm.

In my opinion, this algorithm is not acceptable, however, short of
asking the user to tell the operating system what the block size is, I
have not yet thought of a better solution.

I believe it should be a MUST requirement for SCSI devices to properly
report medium information.

