kenyee@ksr.com (Ken Yee) (01/09/91)
Thanks to all who replied...
Ken (kenyee@ksr.com)
SUMMARY OF SCSI CHIP REPLIES
============================
> - 5 Mbytes/sec Sync (using only an offset of 8...NCR's 53C90 needs
> an offset of 16 for this speed).
So what if the offset is 16? What difference does that make to total
throughput? And you are missing what is by far the most important
factor, which is command setup time. NCR's latest chips have lots of
intelligence for command setup times in 100 usec range, I think.
The 6250 is quite old fashioned in this regard. And I would check with
the pc532 designer, George Scolaro, about the bugs in the 6250. He is
quite pissed off with their technical support and announced he won't
be designing with any Adaptec chips anymore.
A. Lester Buck buck@siswat.lonestar.org ...!uhnix1!lobster!siswat!buck
---------------------------------------------------------------------------
Ken,
We have been very happy designing in the NCR 53C90A ("A" is very significant)
into our products. We build high resolution film printers (a SCSI target
peripheral) where raw throughput is very important (raster images often exceed
100 Megabytes). This chip seems to be very compatible with all of the hosts our
customers have connected it to. We had one minor problem when connecting
to some versions of SGI's operating system, but it turned out to be their
bug, not ours.
As far as speed is concerned, we have had more headaches trying to find a
decent off the shelf DMA controller to design in. The DMA controllers that
are available all seem to be slower than any of the SCSI chips out there
anyways. One promising candidate is something from AT&T (I forget the
part number), but they claim to be discontinuing the part...
Richard A. Keeney Internet: keeney@mgi.com
Management Graphics, Inc. Phone: +1-612-854-1220
1401 East 79th Street, #6 Fax: +1-612-854-6913
Bloomington, MN, 55425
U.S.A.
---------------------------------------------------------------------------
I haven't done a complete comparison with all the chips, but
have been using the 53C700 from NCR and it is approx. 3.2 MB/s Async
and 5MB/s raw SYNC to a Sun SPARCstation. The SPARCstation uses the
53C90 chip set. Three issues come to mind: 1) what is the chip on
the other end? (if you are doing both interfaces then you can optimize
that). 2). Are you including software throughput, since this is
significantly lower than the RAW data rates. 3). If you doing your
own interface and a few bucks (~$50 for the 53C700) doesn't matter,
the order of magnitude ease of use of the 53C700 greatly outways
a couple of MB/s especially if 1 and 2 are considered. The 53C700
application we are using requires 0 interrupts and no code (except
for booting the 53C700) on our processor. (~600 bytes of "SCRIPTS"
code for the 53C700).
uunet!sky.com!knicely (Wendell Knicely)
---------------------------------------------------------------------------
The 6250 pretty seriously sucks. The software overhead required to
use it, completely overshadows any high performance claim about how
fast it can drive the bus. Don't use this one.
The NCR 53C700 is by far a better choice. The 53C710, I believe,
adds the SCSI-2 features, including a fast mode.
Speaking purely for myself...
.---------------------------------------------------------------------.
|Paul Hardin hardin@dg-rtp.dg.com |
|Data General Corp., RTP NC {world}!mcnc!rti!xyzzy!hardin |
`---------------------------------------------------------------------'
---------------------------------------------------------------------------
>
> Thanks for the info on the 6250...I presume the WD33C93A is in the same
> class because I've heard that you need your own state machines to control
> their quirky state machines...
>
Yes. The Western Digital is an improvement over the AIC (since
you probably can't get any worse, I've heard it said that if you
could open the chip package, all you'd see would be jumpers connecting
the inputs to the outputs :-), but the NCR is far better still.
uunet!dg-rtp.dg.com!hardin (Paul Hardin)
---------------------------------------------------------------------------
What do you mean by "fastest"? For example, if chip A can do the
data phase faster than chip B, but chip A generates 5 times as many
interrupts as chip B for the complete command, which is really
fastest?
KY> Fastest overall designable throughput
ts@cup.portal.com (Tim W Smith)
---------------------------------------------------------------------------
Actually, if you are design engineer working on an AT bus type implementation
the AIC-6260 is a better choice. It has the AT bus built in and a much better
software interface than the 6250. It does everything the 6250 does plus some
more stuff. It also support FAST SCSI-2.
> Next 6 months: Adaptec with SCSI-2 chips (SCSI compatible connectors).
The AIC-6251 chip is out and also support FAST SCSI-2.
Roy Neese
Adaptec Senior SCSI Applications Engineer
UUCP @ neese@adaptex
uunet!cs.utexas.edu!utacfd!
{nominil,merch,cpe}!adaptex!neese
uunet!mlite!adaptex!neese
---------------------------------------------------------------------------
The NCR 53C710 should be considered in this thread. Especially if
one considers total time to execute a command, rather than just the
time spent in the data phase.
ts@cup.portal.com (Tim W Smith)ts@cup.portal.com (Tim W Smith) (01/11/91)
The cool thing about the 53C710 is not SCSI-2 but the increased power of the scripts. I've not tried this yet, but it seems to me that one could make a script that would monitor an I/O queue that contains something like Unix buf structures, find requests for idle disk drives, figure out what SCSI command to do, do the command, update status fields in the buf structure, place it on a completion queue, and then interrupt the CPU. The driver interrupt routine would then only have to deal with telling the kernel that I/O has completed. Basically the disk driver would only have an interrupt routine! Tim Smith
feustel@netcom.UUCP (David Feustel) (01/12/91)
I understand that NCR has a development board with the '710 chip on it. Does anyone have info on availability and price? Thanks. -- David Feustel, 1930 Curdes Ave, Fort Wayne, IN 46805, (219) 482-9631 EMAIL: netcom.uucp