Very Computer

I would love to get rid of my future domain 1680 card since I now have
a more capable bus.  I have read various horror stories, but not much
substance about NCR based scsi controllers.

First, since my motherboard has no socket for a NCR chip, I need a whole
card including BIOS (so scsi disk will work with DOS).  Who makes such
beasts?  I'd rather not call salesmen on this because they will just
say "yes" to all my questions so they can make a sale.

Second, the ASUS card is $100.  That is cheaper than many cheapo ISA
cards.  Does this mean it is programmed IO?  

As of now, adaptec isn't an option for me because of their nondisclosure
crap.  Buslogic is an option, but they are expensive, and I want to see
if they are worth three times the NCR price.  Everyone else is not an
option because they do not have linux drivers.

John

Of course not - the NCR53c8xx series only come in surface mount
packages.

Quote:>I need a whole
>card including BIOS (so scsi disk will work with DOS).  

Not necessarily.  Many PCI systems include the NCR SDMS BIOS code
in the system BIOS, although they don't include the chip.  It would
be worthwhile doing a

    strings /dev/mem | grep -i ncr

to see what your memory looks like.

Quote:>Who makes such beasts?  

If you need one with a BIOS chip, Nextor's model 93 is the only one I'm
aware of, unless you want to go with an 825 based board (FAST+WIDE,
requires two minor patches to make it work under Linux).

If not, ASUS, Intel, Portland, and many others make non-BIOS
equipped boards.

Quote:>Second, the ASUS card is $100.  That is cheaper than many cheapo ISA
>cards.  

Actually, you can find non-BIOS versions of the boards for about
$70.  Given the fact that many BIOS vendors (ie, Intel) who lacked
NCR SDMS support for their boards have upgraded their newer ROM
images to included it, this _might_ be the way to go (some of
the ROM vendors can't seem to get their act together,
and alternate between working and broken releases).  Also, any
problems caused by flakey protected mode PCI BIOSes should be
fixed if you upgrade.

The Nextor boards tend to run about $140ish.

Quote:>Does this mean it is programmed IO?  

Nope,  they're busmastering.  Architecturally, they're similar
to the Adaptec AIC-7770/7870 chips used on the 2xxx series, except they
execute more complicated instructions from main memory rather than
simpler ones from on-chip RAM, context switch in software rather than
hardware, and have other minor differences.

Quote:>As of now, adaptec isn't an option for me because of their nondisclosure
>crap.  

Note that Adaptec doesn't have an NDA requirement on the hardware
docs as is generally believed, although there is one on the
downloadable firmware's interface.

However, some of their tech support staff will lie to you about that fact and
refuse to forward your calls to some one who knows what's going on.

IMHO, that's still plenty of reason to avoid Adaptec.

Quote:>Buslogic is an option, but they are expensive, and I want to see
>if they are worth three times the NCR price.  

The Linux Buslogic drivers are a bit more mature.  If you want top
performance from _multiple_ devices accessed at the same time
under Linux, right now instead of whenever I get arround to it (I need to
finish debugging the new save/restore pointers code so the right
values get restored after a SCSI context switch) you'll be happier with
the Buslogic boards.

People've had a lot of problems with the NCR boards, the
vast majority of which break down into general PCI configuration/bug
things.  I've seen the same sorts of reports in large numbers with
the Buslogic boards as well - you'll have it if you have broken
PCI devices or a broken mainboard no matter what you use.

--
Since our leaders won't respect The Constitution, the highest law of our
country, you can't expect them to obey lesser laws of any country.
Boycott the United States until this changes.

Are these the kind of patches for things that nobody realized were
broken until now and will be included in a kernel RSN?  Or the kind
that you have to apply yourself and give rise to catch-22 problems
installing Linux on a new machine?

Quote:>>Buslogic is an option, but they are expensive, and I want to see
>>if they are worth three times the NCR price.  

>The Linux Buslogic drivers are a bit more mature.  If you want top
>performance from _multiple_ devices accessed at the same time
>under Linux, right now instead of whenever I get arround to it (I need to
>finish debugging the new save/restore pointers code so the right
>values get restored after a SCSI context switch) you'll be happier with
>the Buslogic boards.

Is this because of the hardware or the drivers?  Will the NCR boards be as
fast as the Buslogic ones as soon as your drivers get more mature? Or are
the Buslogic boards inherently better (why the price difference anyway)?
--

The kind that you have to install yourself, giving rise to the
catch-22 problem.  I don't have an '825 board, and was unable
to test the code with the '825 so I didn't catch the omission of
an

        '825'

from two switch statements and a left over 2 inside an array that
should have been dimensioned implicitly by the number of initializers.

Quote:>>>Buslogic is an option, but they are expensive, and I want to see
>>>if they are worth three times the NCR price.  

>>The Linux Buslogic drivers are a bit more mature.  If you want top
>>performance from _multiple_ devices accessed at the same time
>>under Linux, right now instead of whenever I get arround to it (I need to
>>finish debugging the new save/restore pointers code so the right
>>values get restored after a SCSI context switch) you'll be happier with
>>the Buslogic boards.

>Is this because of the hardware or the drivers?  

It's entirely a software problem, mostly due to a lack of time.  Basically,
having what worked of the driver stable by the publication of the July
issue of iX Multiuser Multitasking Magazine and lecture notes for
Heidelberg took precedence over getting that working correctly.

I spent a day getting the seriously broken context switching
code fixed, but haven't had a chance to fix the current flaw in
the save/restore pointers code.

(FYI, Stefan Esser has it working fine under his Free BSD driver)

Quote:>Will the NCR boards be as fast as the Buslogic ones as soon as your
>drivers get more mature?

Throughput should be somewhat higher, although CPU usage should be
slightly worse with the NCR boards (More complicated structures
are generated, taking more host CPU cycles).

Quote:>Or are the Buslogic boards inherently better

If you want Adaptec 154x compatability for some reason (Ie, you
need to run an old copy of Xenix), yes.  

Also, the Buslogic boards have onboard active termination where as
the NCR boards I've  seen have been passively terminated.  IMHO,
this isn't really an issue, since you're still looking at a factor
of three price disparity after picking up an active terminator.

Quote:>(why the price difference anyway)?

1.  Parts count

2.  As far as I know, the Buslogic 946 board is unique in that it's
        the only Adaptec 154x compatable PCI board.

3.  People have allways payed $200-$300 for
        bus mastering SCSI controllers.  If they're still
        willing to pay that price, why should Buslogic
        lower their prices?

4.  Multiple sources.  
--
Since our leaders won't respect The Constitution, the highest law of our
country, you can't expect them to obey lesser laws of any country.
Boycott the United States until this changes.