SUMMARY: High Performance VME SCSI Controller ?

From: Jiri Dvorak (dvorak@iam.unibe.ch)
Date: Wed Jan 22 1992 - 15:38:41 CST


I asked:

> With all the new high performance (10MB/sec) SCSI disks available, the need
> for an adequate Controller arises. The standard controllers (Sparcstations
> and Sparcserver300 / 400) are limited to 5MB synchronous mode (I think).
>
> When a 10MB/sec disk is connected to a max. 5MB/sec controller, is it
> possible to run the disk in synchronous mode at 5MB/sec, or will it only
> do async mode? Are there any problems running such disks on slow controllers?
 

First, the assumption that Sun's Controllers for the Sparcsystems 300 and
400 do work sychrounous is WRONG, as some have pointed out. These controllers
are asych only.

Apart from that, a controller and a disk run synch at the slower of the
two rates, i.e. 5MB in the case of a 5MB controller and 10MB disk. There
should be no problems with such a configuration.

I also asked about a specific VME controller:

> I have an offer for a Interphase Cougar VME SCSI-2 Controller,with barely more
> information than just the price. This controller should make 10MB/sec.
>
> Any experiences with this controller?
No reports on that. There are contradictory statements as to whether this
controller is released or not (see below). My sales rep insists on a immediate
availability.

> Do I need an special driver for this controller?
Yes.

> Will it also run in an 600MP?
Yes, with a special driver.

The answers I received contain much vaulable information. I've appended them
to this message.

Many thanks to:
<keith@lgc.com>
<jon@delta.ee.ufl.edu>
<stern@sunne.East.Sun.com>
<Gerhard.Holzer@rcvie.co.at>
<kevin@toad.com>

Jiri Dvorak dvorak@iam.unibe.ch
Institute of Computer Science
University of Berne
Switzerland

**************************included messages********************************
From: " (Gerhard Holzer)" <Gerhard.Holzer@rcvie.co.at>

Are you sure that Sparcserver 400 with standard(!) controllers DO SYNC. SCSI ???
I think they do NOT !
At least ours do not. (OS 4.1.1)
Also SS1 {4/60} can't run SCSI in sync. mode, only async.
SS1+ (note the +) {4/65}, IPCs {4/40} and SS2 {4/75} can do sync. SCSI,
I was told by Sun Munic.
IPC's have a faster internal bus therefore they ran ~5MB and SS[12] ~4-4.xMB.
Which is also my experience.

*******
From: " (Kevin Sheehan {Consulting Poster Child})" <kevin@toad.com>

Sync SCSI involves negotiating for a speed at which both controllers agree
to run. SO the answer is: They will run sync, at the slower of the two
rates the devices care to pick.

*******
From: " (Hal Stern - NE Area Systems Engineer)" <stern@sunne.East.Sun.com>

the 5 Mb/sec limit is the sync tranfer rate limit
(tested limit, that is) on the NCR scsi chip used
in the desktop sparc systems. if you want to
use a VME board in a server, the server can
certainly handle the high throughput (sun quotes
16 Mbyte/sec on the VME bus on a 600MP).

if you want to install the board in a 600MP,
you will almost certainly need a new driver for it.
the 600MP uses the SCSA architecture, rather than
the older SCSI device driver structure, so the
low-level host adaptor driver must conform to
SCSA interfaces.

on a 4/3xx, you should get close to 10 Mbyte/sec,
and you can probably use whatever driver comes with
the board.

*******
>From jon@delta.ee.ufl.edu Tue Jan 21 04:24:39 1992

You're talking about two different kinds of SCSI. Here's the lowdown.

There exist three supported widths for SCSI buses: eight, sixteen, and
thirty-two bits. Eight bit wide SCSI is the most common. The wider versions
came into existence with the advent of SCSI-2.

There exist two modes of operation of the bus: synchronous and asynchronous.
The asynchronous mode of operation has existed since the early days (SCSI-1),
while the synchronous mode of operation is a feature of SCSI-2.

There exist two electrical interfaces to SCSI devices: single-ended and
differential. In synchronous operation, single-ended devices are limited
(in practice) to five million transfers per second, with the actual data
rate then determined by the width of the bus. In contrast, differential
devices are limited to ten million transfers per second, again with the
actual data rate determined by the width of the bus.

Using single-ended transmission (synchronous mode) you can only achieve
five, ten, and twenty megabytes per second with eight, sixteen, and
thirty-two bit buses, respectively.

Using differential transmission you can achieve ten, twenty, and forty
megabytes per second with eight, sixteen, and thirty-two bit buses, respectively.

Why is all of this important? Because the single-ended and differential
devices are not electrically compatible. The board that you're looking
at is probably an eight-bit wide differential SCSI board: wide SCSI is not
common. Thus if you purchase this board and it is in fact a differential
SCSI board then you will not be able to easily get disks, tape drives, etc...
While differential disks, tapes, etc... exist, they are typically high
performance and very expensive -- you don't get the savings of using
the commodity single-ended drives and tape units. Indeed, many products
are not made in differential configurations.

[[According to the offer I have, the board is available either in differential
  or single-ended SCSI --J.Dvorak]]

If you do, in fact, want to connect high performance disk to your machine
then differential is the way to go. This sort of controller will allow you
to connect disks such as the Seagate Sabre series.

*******
>From keith@lgc.com Tue Jan 21 01:54:59 1992
>
> With all the new high performance (10MB/sec) SCSI disks available, the need
> for an adequate Controller arises. The standard controllers (Sparcstations
> and Sparcserver300 / 400) are limited to 5MB synchronous mode (I think).

Sun's VME controllers for the 300 and 400 systems are async!
Sun's SBus SCSI controllers are sync.
The combo SCSI/Ethernet SBus card for the 600 series is sync and 5MB/s.

>
> When a 10MB/sec disk is connected to a max. 5MB/sec controller, is it
> possible to run the disk in synchronous mode at 5MB/sec, or will it only
> do async mode? Are there any problems running such disks on slow controllers?
>

The controller will limit the rate to 5MB/s. The disk should work with no
problem. If you ran this disk on a async controller, it would kick into
async mode for the controller.

> I have an offer for a Interphase Cougar VME SCSI-2 Controller, with barely more
> information than just the price. This controller should make 10MB/sec.
>
> Any experiences with this controller? Do I need an special driver for this
> controller? Will it also run in an 600MP?

Currently Interphase only has the Jaguar VME SCSI controller. The cougar is
not yet released (so my source at Interphase tells me). It will be in the
second quarter '92. My pholosiphy is to let a new product get tested well
by a big OEM so the bugs get worked out before I get it. I usually give
2-3 months for this.

The Jaguar is not SCSI-2 fast, just synchronous.

We have been using the Jaguar for over a year on 370 and 470 platforms with
no problems. It far exceeds the throughput of any of Sun's controllers for
these platforms.

Yes, you do need a special driver for this and it is supplied with the
controller.

We have benchmarked the Jaguar on the 670 system and the combo SCSI/Ethernet
card outperformed it at less cost. I am not sure of the Jaguar
price (approx $2500), the combo card cost $1600 (Sun List).

I doubt very seriously if you see 10MB/s from any controller/disk combination
for a while. We see 3.5MB/s on reads in the perfect environment, and this is
a filesystem read as opposed to a raw device read. 10MB/s is a good rate
from an expensive disk array.

Fast SCSI by spec is 5MB/s rate.

As quoted from the Draft SCSI-2 spec:
"When devices negotiate a synchronous data transfer period of less than 200ns
they are said to be using 'fast synchronous data transfers'"

Also, the spec defines the fastest rate for sync transfer is 10MB/s.
This will happen with fast/wide SCSI. Wide SCSI is a 16 bit bus as opposed to
the current 8 bit bus. And there will be a cable change for wide SCSI.

As I understand it, wide SCSI implementation is of great controversy at this
time. The committees can't seem to agree. This is all street talk.

Currently, several vendors offer what they call SCSI-2. Well this can mean
different things between vendors. The spec is not actually a finished
product. So vendors implement some features which are fairly solid in the
draft spec, such as fast SCSI.

I guess the wrap up here is:

        I do recommend the Interphase Jaguar for 300 400 series Suns. An
        excellent controller. Also a daughter board can be installed to
        give 14 SCSI IDs from one VME slot. This is nice.
        You might mention to interphase, the type of system and OS you
        are running. They have a driver install tape for SunOS 4.1.2 now also.

        I recommend the SBus SCSI/Ethernet card for 600 series. The fastest
        thing we have tested.

        Beware of disk/controller vendors.



This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:06:34 CDT