Path: utzoo!attcan!uunet!van-bc!root
From: lphillips@lpami.wimsey.bc.ca (Larry Phillips)
Newsgroups: comp.sys.amiga.tech
Subject: Re: ESDI drives?
Message-ID: <2084@van-bc.UUCP>
Date: 1 Jan 89 00:10:14 GMT
Sender: root@van-bc.UUCP
Lines: 111

In <12995@cup.portal.com>, thad@cup.portal.com (Thad P Floryan) writes:
> Re: Larry Phillips' comments to my answer to Tony Sumrall's question about
> ESDI on the Amiga ...

>Surprisingly, I get over 250KBytes/second from a Maxtor XT-2190 connected to
>an Adaptec 4000A SCSI<=>ST506/ST412 on the Amiga using 1.3 with FFS (and the
>Supra 4x4 interface).  This performance exceeds that published for most of
>the "modern" systems that are reviewed in, for example, UNIX REVIEW (ref.
>Compaq 386, PS/2-80, certain Sun-3, etc.)

Not overly surprising. The Amiga's FFS is quite quick, and is part of a real
time OS. My current drives, a 40 meg Atasi and a 70 meg Fujistsu, running off
the SCSI port on a 2090A through two Adaptec 4000As, (two, because CBM does not
provide support for LUNs in their driver yet) get 290K and 375K/second
respectively, when they aren't fragmented badly (DiskPerfA, 32K buffer read
figures).  These same drives, running off the ST506 ports would probably get
significantly lower numbers running off the ST506 port on the 2090A.

Since they are ST506 drives, and since they transfer data across the
drive/controller interface at the same rate (5 MBits/sec) regardless of the
controller they are attached to, the difference is obviously in the controller.

>ESDI is a modern, high speed interface, as is SCSI.  "RAW" speed of both a
>modern SCSI and an ESDI interface "should be" around 750KBytes/second, but this
>is not achieved in practice (likely due to an OS and its filesystem)

Remember, ESDI and SCSI are two different 'types' of interface. SCSI is the
part that provides the interface between the host (computer) and the disk
controller.  ESDI provides an interface between the disk controller and the
disk drive.  The 'RAW' speed of an ESDI interface is 10 MBits/sec, and SCSI
speed is dependent upon the handshaking from the host itself, up to a (spec'ed)
maximum of 1.25 MBytes/second, more for extended SCSI, and in fact often more
for standard SCSI, as there is nothing to stop a manufacturer from exceeding
the SCSI max speed spec.

As you can see, the limiting factor in the hardware, assuming every piece of
that hardware runs as fast as it can, is the ESDI interface itself. In
practice, however, the bottleneck, from a hardware point of view, is more often
than not the disk controller itself, and these are definitely not all created
equal.

>The point I wanted to make is that any cost savings (of an ESDI) drive would
>be negated simply by the need to purchase a translator (such as Adaptec's) and
>that one would not see any performance gains; the embedded SCSI drives
>themselves are extremely fast.  Thus, the $200 saved (getting, for example, a
>Maxtor XT-4380 ESDI drive instead of the XT-3380 SCSI drive) are "lost" by the
>need to get a SCSI translator with attendant power cabling, ESDI cabling and
>connectors, etc. (plus having one more "box" in the chain (the translator) as
>a source of potential problems (interface, reliability, whatever...)).

Embedded SCSI drives are also not all created equal. All that has been done
with the embedded SCSI drives is that the functions of the disk controller have
been moved inboard to the drive, effectively 'hiding' the controller/drive
interface completely. The controller/drive interface can be anything from ST506
(probably never dne, but I don't know for sure), to ESDI (fairly common), to a
prorietary and completely non-standard interface running at a higher data rate
than ESDI (I suspect a few of the fastest drives do it this way).

You cite the case of a Maxtor XT-4380 ESDI vs.  an XT-3380 SCSI drive, neither
of which I am familiar with, but can make some general comments about. It's a
fairly safe assumption that drives from the same company, with the same
mechanism, and having the same capacity, one being embedded SCSI and the other
being ESDI, are both operating with the same ESDI controller/drive interface
(it doesn't make economic sense to do it otherwise). So.... why the difference
in performance? In a word, the controller, or as you called it, the SCSI:ESDI
'translator'. To compare the one embedded in the SCSI drive with the Adaptec is
a valid comparison, but it only serves to compare two controllers. Believe me,
there _are_ faster controllers out there. The Emulex MD-21 comes immediately to
mind, and there are probably others.

The real beauty of embedded SCSI is that a manufacturer does not have to stick
with any controller/drive interface, and can do things out there in the way
that will net the best performance. I have always wanted to see, for example,
an 8 data head drive that stores 1 bit position per track, making the sectors
1/8 the length (in time of rotation), and of course coming in 8 times faster.
With embedded SCSI, this is possible, and may already be done in one or more
drives. Without the need for compatibility with other controllers, nifty things
can be done.

>Most of the people I know who've added ESDI drives to their Amigas have done
>so solely because either "the drives were available at work" or "I got the
>drive on an indefinite loan."  For THOSE people, it did make economic sense
>to use the ESDI drive.  To date, I'm not aware of anyone making a "native"
>ESDI interface for the Amiga and, due to SCSI's other advantages, I don't
>foresee anyone doing so on a commercial basis (making an ESDI interface,that is)

I agree wholeheartedly with your cost evaluations, which I guess was the main
thrust of your posting, but wanted to correct the view of SCSI vs.  ESDI that
it presented, as shown by this paragraph.  Since SCSI and ESDI 'look at' two
different parts of the chain between the host and drive, they are just not
comparable.  SCSI does not preclude ESDI, even when a drive has embedded SCSI.
If by 'native' ESDI interface, you refer to something similar to the ST506
ports on a 2090A (or to something like the XT-style ST506 controllers), then it
is quite likely that one will never be produced.  If one were produced though,
it could theoretically transfer data at the full rated speed of the ESDI
interface, or even have a large buffer for caching data (as do many embedded
SCSI drives).  In short, it could perform at the same raw data rate as an
embedded SCSI drive that happens to use ESDI in the controller/drive interface.

>Thad Floryan [ thad@cup.portal.com (OR) ..!sun!portal!cup.portal.com!thad ]

-larry [ somewhre in the wilds of the Great White Frozen North ]

--
"Intelligent CPU?  I thought you said Intel CPU!" 
        -Anonymous IBM designer-
+----------------------------------------------------------------------+ 
|   //   Larry Phillips                                                |
| \X/    lphillips@lpami.wimsey.bc.ca or uunet!van-bc!lpami!lphillips  |
|        COMPUSERVE: 76703,4322                                        |
+----------------------------------------------------------------------+