X3T9.2/93-087 r0
ENGINEERING MEMO 1047  

To:       SCSI RAID Working Group
From:     Gary M. Watson, ADA Project Manager
Date:     3 May 1993

Subj:     RAID Issues From Trimm's Perspective
 

Purpose of RAID Working Group

I get the feeling from reading the minutes of the WG meetings that
your focus is rather narrow, in that you seem to only be talking
about SCSI to SCSI RAID controllers, e.g., the Digi-Data and NCR
approaches.  I can understand why these pop into mind as the most
obvious devices to discuss, but as a percentage of SCSI based RAID
solutions in the field, they may well be on the decline in favor of
software drivers for standard SCSI host adaptors, e.g., CorelRAID,
Integra, 1776, Chantal, Veritas, etc.  Is the WG planning to
address these approaches?  I, for one, think it should.  There is
a third approach that may also be on the rise, that is, embedded
RAID host adaptors, such as Mylex and DPT.  These devices in
particular could stand a little less vendor uniqueness in them.


Integral RAID Drivers

Long term, I see most of the common operating systems offering
software based RAID solutions as part of their standard offerings. 
This reduces user support headaches when the operating systems are
updated.  Someone told me that Windows NT will have native RAID
support right out of the chute.  I don't know what effect this will
have on the business of the SCSI-to-SCSI RAID people, but it can't
be too good.  They will likely have to content themselves with
providing RAID to older platforms that do not have native RAID
support, have no place to plug in additional host adaptors (like
Apple Mac, some Sun systems), or whose RAID solutions are decidedly
inferior.  From a performance standpoint, it appears that the
native RAID solutions are on a par performance-wise with the SCSI-
to-SCSI solutions, yet they usually cost little more per megabyte
than ordinary, non-redundant disk subsystems.


Hot Swap Electronic Issues

One of the key advantages of RAID technology in the context of
multi-user computers or file servers is that with the use of hot
swappable disk drives, power supplies, and fans, the system can
remain "available" even during servicing.  However, as far as I can
see, the SCSI spec is silent on the topic of inserting or removing
disk drives on a live and active SCSI bus.  Quite a number of
people do it, and with proper design it seems to work, but it would
set a lot of minds at ease if the SCSI committee could "bless" the
concept and establish design and/or testing criteria.  Or,
alternately, condemn the idea and offer a workable bus isolation
circuit that could be economically implemented.  Also, perhaps
specify the power up and down behavior of a peripheral's SCSI bus
transceivers (e.g. it can't spuriously assert a bus signal.)


Enclosure Statuses and Controls

There are more types of failures in a RAID system than just disk
drives.  Power supplies can fail.  Redundant load sharing boards
can report a problem.  Fan stall detectors can indicate a potential
thermal problem, and cabinet thermistors can detect an actual
thermal problem (such as loss of air conditioning.)  Further, there
are indicators on a disk enclosure that theoretically could be used
by the RAID driver to indicate which disk module has failed and
needs to be swapped, and our company plans to offer mechanical
interlocks on our enclosures that the remote system can use to
prevent or allow a module's removal.  Presently we plan to allow
these controls through a RS-232C serial port, but they could be
controlled via some sort of SCSI model as well.  Preferably, one
SCSI target ID would control all cabinets of disk drives if there
happened to be more than one.


Other Random Topics

I would like to see the RAID WG think about redundant spindle sync
for RAID 3 arrays, as without it their performance can suffer to
the point that the array is useless, and this must also have hot
swap capability.  I would also like to see the WG include IPI in
its discussions, at least in general terms.  Can you hot swap IPI?

I may try to attend the May 17 meeting if my schedule permits.

Gary Watson
Trimm Industries
11949 Sherman Road
North Hollywood CA 91605
(800) 423-2024  fax (818) 503-0438