Pentium Pro 200 - 256k cache vs 512k & 1MB cache

In article ,
lid says...
Keith R. Williams wrote:
I cannot understand why you need a PC/AT for *anything*.

The PC/AT and clones up thru the 486 vintage had superior
ISA busses running at 6-8 MHz (4-5.33 MBy/s). Later Pentiums
with PCI busses have Southbridges which slow ISA effectively
down to 1 MB/s.

More or less true. 1MB/s is a tad on the negative side of the
situation though.

This also affects interrupt latency since
the PIC lives off ISA. Better to use an APIC-aware OS.

The PIC/APIC and OS have little to do with it. Only Intel
supported APIC and that was a P6 thing.

But if you have a demanding [real-time] app like an instrument
controller card working on the ancient ISA bus, you may not
be able to upgrade past a 486.

Sheesh! If this is your problem extend a ATA port to an ISA
backplane. Ok, you're going to have limited addressing, but if
you're locked into such antiques, you certainly have my pity!

--
Keith

In article , fammacd=!
says...
On Sun, 7 Sep 2003 20:13:41 -0400, Keith R. Williams
wrote:

In article , fammacd=!
says...
It also seems that they often decide that
they prefer to stay here and "convert" to a resident alien status.

This can happen, though it's difficult. IMHO, should be
impossible!

I'm not sure exactly what or why you want to be impossible. At least they
have experienced the culture, work/social environment and decide they like
it and would rather live in the U.S... not that different from every other
immigrant??

They should go home and re-apply for entry like everyone else.
In many cases their *government* funded their education and
expect a return. In *all* cases the US government doesn't expect
them to stay (hence no "green card").

snip

Ok, let me put it another way... If I could pass your test, it
is silly. Yes, I understand computer architecture (some are
beyond my interest:concentration). If you're looking for
candidates to *program*, I'd think you'd be looking for those who
understand file-syetems, data-structures, C++, and all that un-
Holy jazz.

Quite frankly no - the ability to do the "painting by numbers";-) job of
C++ is not *that* important to us - they can learn that as needed. IMO
real programming skills, without a debugging IDE to bounce around in is a
somewhat different asset.

I'm not sure what's left there. Color me confused. From past
conversations (please correct me if I'm wrong) you do analytical
programming, yet require someone who knows how to trace a bit
through a processor?

BTW this PC AT is one which I understand has 3.5" diskette support in the
BIOS but the IBM Setup utility did not cater for it so we have to use 5.25"
diskettes.

I cannot understand why you need a PC/AT for *anything*.

AFAIK the AT/370 board, VM/PC may not work in anything else but a real IBM
PC/AT with IBM-DOS. The memory is an Intel Above Board strapped for
Extended Memory.

There *are* PCI versions of this thing (that knock the crap outta
the ISA ones). Your use of the name sorta confused me. I'm not
so sure what the relevance of this thing is to the real world is
though. No one does anything other than business and legacy
computing on 370s anymore, do they?

I've searched high and low for a later Setup utility which
includes the 3.5" support... to no avail.

Huh? This is not part of the hardware, nor BIOS. Even so, it's
rather easy to add boot devices into the I/O channel. Perhaps
I'm missing a key TTB (technical-tid-bit) here.

It's easy if you have access to programming of the CMOS to put the device
code into the CMOS table, i.e. the Setup utility. From what I've read,
early PC/ATs had no possibility of 3.5" diskette drive because the BIOS did
not contain the code for accessing it - the later ones, which we have, did.
This is DOS based remember and all devices are accessed through the BIOS
code.

You're right, the BIOS in an AT *hardware* didn't accept 3.5"
drives, though an adapter card with their own BIOS did. Remember
the AT (and XT, IIRC) allowed BIOS to be inserted in the I/O
channel.

--
Keith

In article , fammacd=!
says...
On Mon, 08 Sep 2003 05:45:22 GMT, Tony Hill
wrote:

On Mon, 08 Sep 2003 03:52:34 GMT, Robert Redelmeier
wrote:
But if you have a demanding [real-time] app like an instrument
controller card working on the ancient ISA bus, you may not
be able to upgrade past a 486.

shudder An important real-time app that depends on an ISA card?!
Geez, you've got to be some sort of masochist to subject yourself to
such at thing! Next thing you're going to tell me that it runs under
Win95! :

You'd be surprised - there's a whole boatload of old industrial controller
systems out there still doing SCADA off the ISA bus. In many cases the
controller cards were never made available for PCI or the cost for the card
alone would be as much as several PCs.

Didn't you ever read the story about a company which, in the process of
reclaiming its mainframe glasshouse for other use as it transitioned to PCs
and distributed servers, tore down a wall to find a 20year-old Honeywell
(200 ?) system which was *still* controlling the air-conditionng system for
the entire building. It never broke so it got forgotten.

I've heard the same story about a PS/2 running OS/2, a PDP-8
running (&deity knows what), and a grad-student locked in the
basement. Seems like Snopes time again. ;-)

--
Keith

In article ,
says...
On Mon, 08 Sep 2003 12:40:20 GMT, Robert Redelmeier
wrote:

Tony Hill wrote:
shudder An important real-time app that depends on an ISA card?!
Geez, you've got to be some sort of masochist to subject yourself to
such at thing! Next thing you're going to tell me that it runs under
Win95! :

Win95? Definitely not! These machines were installed
_way_ before that and run something like MS-DOS 3.3 .

No masochism is involved. These machines were installed in
the late 80s and early 90s and just simply work. Occasional
dustblowings and rare disk failures.

It's easy to justify the first computerization of a task.
It's much tougher to justify computer upgrades.

-- Robert

On a similar vein, the Electrical shop where I work could not justify
upgrading (again) from a Celeron 666 ? (ISA card) to a P4 - 2.4, as
the PCI card that connects to all the PLC's in the plant was $
2,000.00 + each. The computer / networking tech said that nobody makes
a MB for a P4 with ISA slots.

On a bet, if I could upgrade you for less than a new PCI card, would
you buy it ? Guess who won :-)

Although the outside world communicates on the ISA bus, the inside
world, searching ladder logic, databases, or whatever, saw a good
speed increase. At $ 200.00 + per minute of downtime, yes Tony,
there is still a need for fast computers, and an ISA bus :-)

Let me put it another way; $2000 is cheap, when you're business
comes to a halt because there are no more ISA cards and your
antique system just went casters-up. You'd better find a way to
change, and *FAST*. ISA is just as dead as the rest of the
dinos! ...and good RIDDANCE! OTOH, perhaps there is real money
in them hills, Hey Tony?

I have no idea what the problem is PCI/ISA "bridges" (like the
PCI9052) are out there. The redesign really amounts to some
driver work and a spin of the widget. Me thinks people protest
too much.

Running W2k, system idle ~ 95 % most times. Now if I can get these
cheap farts to buy me a 19" monitor (or two).

Since they're a cheap too...

--
Keith

On Mon, 8 Sep 2003 21:39:42 -0400, Keith R. Williams
wrote:

In article , fammacd=!
says...
On Sun, 7 Sep 2003 20:13:41 -0400, Keith R. Williams
wrote:

I'm not sure exactly what or why you want to be impossible. At least they
have experienced the culture, work/social environment and decide they like
it and would rather live in the U.S... not that different from every other
immigrant??

They should go home and re-apply for entry like everyone else.
In many cases their *government* funded their education and
expect a return. In *all* cases the US government doesn't expect
them to stay (hence no "green card").

There are also a substantial number who are at a State funded university,
paying in-State tuition, because they have an uncle/auntie here who is a
citizen.:-) My son saw this at Rutgers. I see your point though.

snip

Ok, let me put it another way... If I could pass your test, it
is silly. Yes, I understand computer architecture (some are
beyond my interest:concentration). If you're looking for
candidates to *program*, I'd think you'd be looking for those who
understand file-syetems, data-structures, C++, and all that un-
Holy jazz.

Quite frankly no - the ability to do the "painting by numbers";-) job of
C++ is not *that* important to us - they can learn that as needed. IMO
real programming skills, without a debugging IDE to bounce around in is a
somewhat different asset.

I'm not sure what's left there. Color me confused. From past
conversations (please correct me if I'm wrong) you do analytical
programming, yet require someone who knows how to trace a bit
through a processor?

Well, not quite "trace a bit" but we have quite a mix of stuff, from
numerically intensive to some where an engineer with good programming
skills is ideal. We're no longer actively looking for anybody right now.

AFAIK the AT/370 board, VM/PC may not work in anything else but a real IBM
PC/AT with IBM-DOS. The memory is an Intel Above Board strapped for
Extended Memory.

There *are* PCI versions of this thing (that knock the crap outta
the ISA ones). Your use of the name sorta confused me. I'm not
so sure what the relevance of this thing is to the real world is
though. No one does anything other than business and legacy
computing on 370s anymore, do they?

Yeah IIRC you mentioned modern versions of the AT/370 before. Do they
still used hacked 68020s? I think you have to own a mainframe, which we
don't have, to get access to them - we got the existing ones through the
backdoor... of an independent service bureau, which probably broke all
kinds of official "rules". They really aren't used that much anymore...
apart from that one thing which is not in any way performance realted.

Rgds, George Macdonald

"Just because they're paranoid doesn't mean you're not psychotic" - Who, me??

Keith R. Williams wrote:
More or less true. 1MB/s is a tad on the negative side of the
situation though.

Thanks for the confirmation. I used some old tools from
c't and get 1.2 MByte/s at best on later ISA busses.

The PIC/APIC and OS have little to do with it.
Only Intel supported APIC and that was a P6 thing.

Uhm ... then why is there K7 code in
/usr/src/linux/arch/i386/kernel/apic.c ?
Don't know (but doubt) Intel dropped it in the P7.

Pray tell, if you want to service more than 200,000 interrupts
per second (the max I've measured on a PC), how would you
do it? The [virtual] 8259 seems a big bottleneck. With APIC,
breaking a million looks possible on GHz+ CPUs.

you're locked into such antiques, you certainly have my pity!

Thank you. Eventually, they become unmaintainable and have
to be upgraded. That's the theory, anyways.

-- Robert

On Tue, 09 Sep 2003 14:18:22 GMT, Robert Redelmeier
wrote:

Keith R. Williams wrote:
More or less true. 1MB/s is a tad on the negative side of the
situation though.

Thanks for the confirmation. I used some old tools from
c't and get 1.2 MByte/s at best on later ISA busses.

The PIC/APIC and OS have little to do with it.
Only Intel supported APIC and that was a P6 thing.

Uhm ... then why is there K7 code in
/usr/src/linux/arch/i386/kernel/apic.c ?
Don't know (but doubt) Intel dropped it in the P7.

Recent NForce2 systems I've used have a BIOS Setup parameter for the APIC
version to use. Isn't MSI dependent on APIC or some kind of, "not
necessarily Intel", similar scheme? I'd thought MSI was where we were
going.

Rgds, George Macdonald

"Just because they're paranoid doesn't mean you're not psychotic" - Who, me??

In article ,
lid says...
Keith R. Williams wrote:
More or less true. 1MB/s is a tad on the negative side of the
situation though.

Thanks for the confirmation. I used some old tools from
c't and get 1.2 MByte/s at best on later ISA busses.

The PCI uses "subtractive addressing" (six PCI clocks) to get
across the ISA bridge. These bridges aren't designed to be all
that high performance. Add in the time for an ISA card to ACK an
address and it's up to eight times ISA/PCI latency. PIO can
easily be in the 1Mbps range. ...usually a bit above (perhaps
2X). ISA DMA is another kettle-o-fish!

The PIC/APIC and OS have little to do with it.
Only Intel supported APIC and that was a P6 thing.

Uhm ... then why is there K7 code in
/usr/src/linux/arch/i386/kernel/apic.c ?
Don't know (but doubt) Intel dropped it in the P7.

Whoop! Whoop! Brain-fart time! Whoop! ...and whoops!

Intel would not license *their* APIC to AMD nor Cyrix. That
doesn't mean the PIC is still on the ISA bus. I guess my brain
isn't working all that well at night.

Pray tell, if you want to service more than 200,000 interrupts
per second (the max I've measured on a PC), how would you
do it? The [virtual] 8259 seems a big bottleneck. With APIC,
breaking a million looks possible on GHz+ CPUs.

See above. :-(

you're locked into such antiques, you certainly have my pity!

Thank you. Eventually, they become unmaintainable and have
to be upgraded. That's the theory, anyways.

If you're using them as a "entrance exam" I hope you're keeping
your HR and legal types informed. I know 30ish years ago
entrance exams were forbidden for engineers (though not
programmers, because they were not "professionals"), but there
were (and are) far better ways around the formal exam. "What
projects have you done, please explain."

--
Keith

On Mon, 8 Sep 2003 22:52:53 -0400, Keith R. Williams
wrote:
In article ,
says...
Although the outside world communicates on the ISA bus, the inside
world, searching ladder logic, databases, or whatever, saw a good
speed increase. At $ 200.00 + per minute of downtime, yes Tony,
there is still a need for fast computers, and an ISA bus :-)

Let me put it another way; $2000 is cheap, when you're business
comes to a halt because there are no more ISA cards and your
antique system just went casters-up. You'd better find a way to
change, and *FAST*. ISA is just as dead as the rest of the
dinos! ...and good RIDDANCE! OTOH, perhaps there is real money
in them hills, Hey Tony?

I have no idea what the problem is PCI/ISA "bridges" (like the
PCI9052) are out there. The redesign really amounts to some
driver work and a spin of the widget. Me thinks people protest
too much.

There's always money in them Hills :

Here's an interesting little board for you:

http://www.supermicro.com/PRODUCT/Mo.../875/P4SCA.htm

Brand new board based off of Intel's i875P chipset, supporting 800MHz
bus speed P4's, but would ya lookie here... not 1, but *THREE* 16-bit
ISA slots! Now this is obviously a real specialty product, but I
guess it shows that there is some sort of demand for ISA slots still.

--------------
Tony Hill
hilla underscore 20 at yahoo dot ca

In article , fammacd=!
says...
On Mon, 8 Sep 2003 21:39:42 -0400, Keith R. Williams
wrote:

In article , fammacd=!
says...
On Sun, 7 Sep 2003 20:13:41 -0400, Keith R. Williams
wrote:

I'm not sure exactly what or why you want to be impossible. At least they
have experienced the culture, work/social environment and decide they like
it and would rather live in the U.S... not that different from every other
immigrant??

They should go home and re-apply for entry like everyone else.
In many cases their *government* funded their education and
expect a return. In *all* cases the US government doesn't expect
them to stay (hence no "green card").

There are also a substantial number who are at a State funded university,
paying in-State tuition, because they have an uncle/auntie here who is a
citizen.:-) My son saw this at Rutgers. I see your point though.

Silly taxpayers. When I was in school I couldn't get *any* help
until I was on my own for *two* years. I was married and on my
own but they still wanted the financial from the parents.

My mother (father died when I was a kid) had cash because she
sold her huge house and was down-sizing, and building a smaller
house (remember, I was married and out of the house) I couldn't
get squat assistance for the two years. No, I'm not saying that
the taxpayers were unfair. Special cases tend to fall through
the cracks. I can't believe taxpayers are dumb enough to take
care of nephews/nieces without similar time issues. Of course
when states give in-state tuition to *ILLEGAL* aliens, I guess
nothing should surprise me.

snip

Quite frankly no - the ability to do the "painting by numbers";-) job of
C++ is not *that* important to us - they can learn that as needed. IMO
real programming skills, without a debugging IDE to bounce around in is a
somewhat different asset.

I'm not sure what's left there. Color me confused. From past
conversations (please correct me if I'm wrong) you do analytical
programming, yet require someone who knows how to trace a bit
through a processor?

Well, not quite "trace a bit" but we have quite a mix of stuff, from
numerically intensive to some where an engineer with good programming
skills is ideal. We're no longer actively looking for anybody right now.

I'm not actively looking either, now. ;-) I simply find it
amazing that you would be looking for these *skills* as a
*requirement*. Perhaps one that understands the innards of a
processor is somehow useful in the real world? ;-)

AFAIK the AT/370 board, VM/PC may not work in anything else but a real IBM
PC/AT with IBM-DOS. The memory is an Intel Above Board strapped for
Extended Memory.

There *are* PCI versions of this thing (that knock the crap outta
the ISA ones). Your use of the name sorta confused me. I'm not
so sure what the relevance of this thing is to the real world is
though. No one does anything other than business and legacy
computing on 370s anymore, do they?

Yeah IIRC you mentioned modern versions of the AT/370 before. Do they
still used hacked 68020s?

Nope, dedicated processors. There was (is?) a group doing such
things. The group was under a fellow, though I haven't heard
from them in a few years.

I think you have to own a mainframe, which we
don't have, to get access to them - we got the existing ones through the
backdoor... of an independent service bureau, which probably broke all
kinds of official "rules". They really aren't used that much anymore...
apart from that one thing which is not in any way performance realted.

I never heard that one needed a "mainframe" to buy one. I know
you have to license VM and all that goes with it. IIRC the
license wasn't all that bad though. YOu might ask on comp.arch.
I'm sure there are several there who know more.

--
Keith