30 Jul
2024
30 Jul
'24
11:58 a.m.
Pretty much, though Windows2000 was a significant redesign over NT4.
I was one of the first users of NT4 at the IEEE Computer Society in the
early 90's. We ran it on TALOS, the 8 CPU NCR 3550 system and it was one
of the first truly SMP systems out there (with dual Microchannel busses
as well). We needed the power because it ran the Computer Society's
Digital Library, a neat set of hacks running on Netscape commerce
server, a SGML-HTML converter called DynaWeb and a custom little program
that I wrote to convert TEK math into gifs to embed into the HTML
documents so the math looked perfect.
For this kind of highly distributed process (call the server to get the
page, Dynaweb to render the text and fonts, and TEK to convert all the
formula to embedded graphics) we really needed all the CPU power we
could get and frequently had the system running 8 CPUs at 70%-80%.
Ultimately we upgraded it to 16 Pentium Pro/200 CPUs and finally to 16
Pentium Pro Overdrives running at 333mhz*
The big problem with NT was that the TCP/IP stack kept crashing. Ran out
of memory even though we had 512mb of memory in this beast. I then
remembered my RSX11M days and how M systems would crash with out of
memory errors with plenty of free memory in RMD. In that case
unterminated RS232 connections were picking up interference and were
sending streams of text to the TT: driver. This would run the system out
of pool and since the pool partition was a fixed region size it would
sink the server.
Worked with Microsoft and sure enough: The issue was system controlled
partitions with the TCP/IP driver being loaded into a partition that had
limited memory. For NT they patched it, but for Windows 2000 they
flattened the former various partitions into what was essentially GEN.
Yes, they were stunned that we figured this out. Yes, there are boot
options in 2000 to fix all this. But Windows 2000 was a re-architected
version of NT that people hated because of the GUI. So they dusted up
the GUI with Windows 98's stuff and released it as XP. Rest was history....
CZ
* Yes, you could run 16 Pentium overdrive chips even though their design
limited SMP to two CPUs. The trick was the NCR 3550's Pentium Pro board
had 4 chips on it but they ran in pairs of two. In other words two
PPro's communicated to a L3 cache controller which then used its' own
glue logic to connect to one side of the dual 256 bit wide memory buses.
So to the CPUs they just saw their neighbor. The other pair of two CPUs
was connected its own level 3 cache that hooked into the second 256 bit
memory bus which went to the 4 memory cards (dual ported cards).
The two MCA channels also attached to the big bus, one per side. Thus it
was possible to have both MCA channels doing DMA at full blast speed to
memory and the system would just take it. We duplexed the drives and got
amazing performance with RAID5 on each bus.
The only trick to this was that processes would rip like hell in cache
if they stayed on their own PPro CPU (running out of L1/L2 cache) and
would run fast if they were within their pair of processors (dump L1/L2,
but L3 was super quick). What slowed you down was if the process was
moved to another board which meant you had to reload L1, L2, AND L3
cache which was a serious time sink. 256 bit memory fetch ahead helped,
but it would cause cache storms on the system as processes started
leaping around on a full system.
The solution was to use process affinity to essentially "carve" the
system up into 8 regions of two processors and have affinity prefer to
keep certain processes (Dynaweb got 8 of the CPUs, TEK got 4, and
Netscape/OS got the last 4) close to their CPUs when possible. But if a
CPU was totally idle and a neighboring cluster was asking for more TEK
math converters the system would start up new processes on the unused
processor (no real cache hit) but not move it off there till it
terminated (most were short lived)
Fascinating times. We really had a lot of fun with that box. In a way,
it was the SMP RSX11M+ I always wanted to play with.
On 7/30/2024 4:42 AM, Dave Wade G4UGM via cctalk wrote:
-----Original
Message-----
From: David Wise via cctalk <cctalk(a)classiccmp.org>
Sent: Tuesday, July 30, 2024 2:01 AM
To: Murray McCullough via cctalk <cctalk(a)classiccmp.org>
Cc: David Wise <d44617665(a)hotmail.com>
Subject: [cctalk] Re: MS-DOS
I think Windows 2000 is NT-based.
Yes it started life as NT5 but at some point in got renamed to 2000 and DEC
Alpha support was dropped. I may have some NT5 Beta CDs in the loft.
Dave
Dave Wise
________________________________
From: Fred Cisin via cctalk <cctalk(a)classiccmp.org>
Sent: Monday, July 29, 2024 5:21 PM
To: Murray McCullough via cctalk <cctalk(a)classiccmp.org>
Cc: Fred Cisin <cisin(a)xenosoft.com>
Subject: [cctalk] Re: MS-DOS
On Mon, 29 Jul 2024, Murray McCullough via cctalk wrote:
chapter in
I had not realized that 43 yrs. ago Microsoft
purchased 86-DOS for
$50,000 - US not Cdn. money. With this purchase the PC industry, IBM's
version thereof, began. I remember using it to do amazing things,
moreso than what 8-bit machines could do!
Ah, but there is so much more to the story, which deserves an entire the history.
More than you wanted to know? : (but even more details available if you
really
want them)
Tim Paterson, of Seattle Computer Products was developing 8086 hardware,
but
CP/M-86 was delayed. So, he wrote a temporary
place-holder to use instead
of
CP/M-86 until CP/M-86 became available. That was
called "QDOS", "Quick
and
Dirty Operating System". Later it became
known as "SCP-DOS" and/or
"86-DOS"
Then came the "culture clash" between IBM and Digital Research (previously
known as "Intergalctic digital Research"). That has been documented
elsewhere; some claim that there was not a culture clash, nor an error.
So, Microsoft (possibly Bill Gates personally) went down the street to
Seattle
Computer Products, and bought an unlimited
license for 86-DOS "that we can
sell to our [un-named] client"
Tim Paterson, who later opened "Falcon Technologies" and Seattle Computer
Products both also retained licenes to be able to sell "the operating
system".
Note that the version was not specified, as to
whether such license would
include rights to sell updated versions; that error (failure to specify
whether
future/derivative products were included) has
been repeated elsewhere (cf.
Apple/Microsoft)
Microsoft also hired Tim Paterson to maintain and update "MS-DOS".
Microsoft sold a license to IBM, where it became PC-DOS.
And, it was available through Lifeboat as "86-DOS"
In August 1981, when the PC (5150) was released, IBM started selling
PC-DOS.
But digital Research was not happy with IBM
selling a copy of their
operating
system.
In those days, selling a copy was legal, if the internal code was not
copied.
(hence the development of "clean-room
reverse engineering") It wasn't
until the
Lotus/Paperback Software (Adam Osborne) lawsuit
that "look and feel"
became
copyrightable.
So, IBM agreed to also sell CP/M-86 IN ADDITION to selling PC-DOS.
. . . and sold UCSD P-System.
But CP/M-86 was STILL not ready, so everybody bought PC-DOS, many of whom
planned to switch to CP/M-86 when it became available.
But, when CP/M-86 was finally ready, the price was $240 vs $40 for PC-DOS.
There are arguments about whether IBM or Digital Research set that price.
Although, if that price was IBM's idea, then why did Digital Research
charge
$240 for copies sold through other sources (such
as Lifeboat)?
Initially MS-DOS and PC-DOS differed only in name and trivial items, such
as
"IO.SYS" and "MSDOS.SYS"
being renamed "IBMBIO.COM" and "IBMDOS.COM"
When changes were made, Microsoft's and IBM's version numbers were
separated.
Thus 1.00 was the same for both
IBM released PC-DOS 1.10, and Microsoft released MS-DOS 1.25
2.00 was the same for both
2.10 VS 2.11 (IBM needed trivial changes to 2.00 to deal with the
excessively
slow Qumetrak 142 disk drives in the PC-Junior
and "portable"
3.00 was the same
3.10, adding network support and the "network redirector for CD-ROMs
3.20 VS 3.21, adding "720K" 3.5" drive support
3.30 VS 3.31, BUT 3.31 was the first to support larger than 32Mebibyte
drives!
4.00 and 4.01 IBM/Microsoft did not provide
third party vendors enough
advanced warning, so Norton Utilities, etc. did not work on 4.00 (NOT
4.00 did not work with Norton Utilities!)
5.00
In 6.00 each company bundled a whole bunch of third party stuff (such as
disk
compression) and each got them from different
sources.
When Microsoft's disk compression was blamed for serious problems caused
by
SMARTDRV, Microsoft released 6.20 (repaired and
reliability improved from
6.00).
Then 6.21 and 6.22 as a result of Microsoft's legal case with Stac
Electronics.
Please note that MS-DOS/PC-DOS ALWAYS had a version number, a period, and
then a TWO DIGIT DECIMAL sub-version number. THAT is what is stored
internally. Thus, 1.10 is stored as ONE.TEN (01h.0Ah), 3.31 is actually
THREE.Thirty-ONE (03h.1Fh), etc.
If there had ever actually been a "1.1" or "3.2", those would have
been
01h.01h
(1.01) and 03h.02h (3.02), etc.
"1.1" was NOT the same as "1.10", nor "3.2" the same as
"3.20", otherwise
VERY
minor changes would be confused with serious
changes, as happened when
some people called 4.01 "four point one".
Later still, Seattle Computer Products was on the rocks. There was some
speculation that AT&T might buy it, to get the DOS license (and not have
to
pay
royalties per copy!). After some legal
animosity, Microsoft did the right
and
smart thing, and bought Seattle Computer
Products, thus closing that
vulnerability.
Windows originally started as an add-on command processor and user
interface
on top of DOS. Windows95 made that invisibly
seamless, so the user never
saw
a DOS prompt without explicitly asking for it.
Windows 95 still contained
DOS
(7.00), but the user never saw it.
Gordon Letwin at Microsoft developed OS/2. But Microsoft sold it off to
IBM,
and it became known as an IBM product.
Microsoft used some key technology from it in developing WindowsNT.
Within Microsoft's offerings, NT competed with non-NT windows, such as
Windows95, Windows98, and Windows2000.
Windows[NT] Vista, XP, 7, 8, 9, 10, 11 continued, and the old Windows was
"deprecated'.
Naming a version after the year it is released is great for sales in the
first
year,
and a serious liability in subsequent years,
unless there is actually
going to be a
new version every year (as automobiles do)
--
Grumpy Ol' Fred cisin(a)xenosoft.com