Re: [WinMac] Re: Re: dual processors on NT Servers

Welch, John C.(jwelch[at]aer.com)
Wed, 06 Oct 1999 10:37:15 -0400

lol...A good Intel flame would need 3 typists and a T-3!

> From: "Daniel L. Schwartz" <expresso@snip.net>
> Reply-To: "The Windows-MacOS cooperation list"<winmac@xerxes.frit.utexas.edu>
> Date: Wed, 06 Oct 1999 00:09:56 -0400
> To: "The Windows-MacOS cooperation list" <winmac@xerxes.frit.utexas.edu>
> Subject: [WinMac] Re: Re: dual processors on NT Servers
>
>
> Uh oh... here we go!
>
> At 11:43 PM 10/5/99 -0400, John wrote:
>> Comments inline.
>>
>> NOTE: Despite what it may look like, this is a Microsoft flame, and not a
>> Dan Flame. Having actually met Dan at MacWorld, I can attest that we both
>> enjoy high-spirited arguments, which are NOT the same as personal attacks.
>
> Agreed... But it also looks like an intel flame (see below).
>
>>> Yes, NT is, and always has been. a fully symmetrical multiprocessor
>>> operating system. Unlike MacOS & NetWare, which are pretty much
>>> asymmetrical, NT's thread scheduler will gladly dispatch tasks to the
>>> available CPU's by which one has the lightest load.
>>>
>>> BTW, along these same lines, this was one of the reasons that NT blew
>>> linux out of the water in the controversial NT-linux shootout in early
>>> summer: The testbed was a 4 CPU Dell box with 4GB of RAM. When the second
>>> round of tests were performed - With RedHat & Microsoft engineers each
>>> optimizing their respective systems - it was discovered that linux didn't
>>> have a multithreaded TCP/IP stack!
>>>
>>
>> unfortunately NT's thread scheduler is a joke, so one thread...i.e. a 2 page
>> word document, can eat 99% of the CPU, which doesn't happen on Unix. NT's
>> scaling also goes in the terlet past 4 processors, whereas Unix and AS/400s
>> scale happily to many more processors, support *real* clustering, and have
>> uptimes of up to 5 years on average for the 400s. But hey, NT beat Linux, so
>> that's all that matters.
>
> Actually, over on the AlphaNT list we've been talking about this very
> issue with the 8 way xeon (rip-off) servers coming out. BTW, everyone but
> Compaq is having all kinds of hell with their 8 way designs using intel
> boards. Only Q designed their own boards; and theirs work.

Isn't it cool that Intel is working it's heiny off to be the sole source of
x86 design specs...with SUCH high quality, why go anywhere else?

>
>
>>> ------
>>>
>>> Just imagine the (hypothetical) men's room conversation at the lab:
>>>
>>> "Let's see, we don't have a multithreaded IP stack. Let's throw this into
>>> the kernel and recompile..."
>>>
>>> "What about regression testing for quality & compatibility before putting
>>> this out there?"
>>>
>>> "Ha Ha! This is Linux - As long as it works on this machine, fine.
>>> Everyone else'll have to edit the kernel to suit their own "needs:" This is
>>> "open source" at it's finest!"
>>
>> As opposed to the NT convo: "Gee, we can fix the umpteen bazillion security
>> and reliability bugs...NAH...instead, let's tie everything into a pseudo -
>> hierarchal(sp?) directory that is buzzword compliant, but won't be really
>> functional for at LEAST 3 service pack..." MS operating systems at THEIR
>> finest.
>
> Well, it took 4 service packs for 3.51 (7 if you count NT 3.5), 3 SP's for
> NT4, and with 65 million (and growing) lines of source code any guesses for
> NT5/W2k?!
>
> Besides, security in NT is fantastic: It's is fully C2 compliant... As
> long as you don't hook it up to a network! :)
>
>>> ------
>>>
>>> Going back to the original NT question, applications don't need to know
>>> anything about the hardware: They are "cradled" by the HAL (Hardware
>>> Abstraction Layer) of NT, and NT will "break up" the application into its'
>>> DLL (Dynamic Link Library) threads and distribute the threads amongst the
>>> CPU's.
>>
>> If the HAL is so good at *cradling* apps, how come it couldn't abstract the
>> processor, and why so much assembly optimizations for Intel, and not MIPS,
>> Alpha, or PowerPC?
>
> Exactly my point: This is (sorta) an *INTEL* problem, not an NT problem.
> But, NT was also ported to i860 and i960. It's that x86 legacy crap that
> requires all of that hand coding... And don't you think Linus has spent
> more than a few sleepless nights using an x86 assembler?
>

Isn't it cool that Intel is still compatible with 8088 assembly code? Geez,
I like my thermometer, but I take it out occasionally!

>>>
>>> [Note that there *is* an exception: old 16 bit apps, which are run in
>>> NTVDM (NT Virtual DOS Machine, an x86 emulator): Normally, NTVDM's run in
>>> their own separate processes; but there are a few cases where the
>>> "DefaultSeparateNTVDM" key has to be set to "No" in order for multiple
>>> legacy apps to communicate between each other. In this case, NTVDM.EXE will
>>> also run on only a single CPU. The obvious answer is to get rid of as much
>>> of the 16 bit crap as possible: On a well-oiled NT Server this shouldn't be
>>> too difficult]
>>
>> Why do I have to start gutting NT to make it well oiled. By the time that's
>> done, I'LL need to be well oiled. Rebel Beer anyone? I never do this with
>> Solaris...gee, good thing NT's SO much easier than Unix.
>
> Aw, come on... All it takes is applying a few .REG files to tweak a few
> keys, and all is fine.

Nah, homey don't play that. If Microsoft can't be bothered to do it right,
why should I? I also don't go in and gut shipped OS extensions on my Macs
routinely. The only time I've violated this was for our webstar box, which
has no AppleTalk or AppleShare extensions running. Security concerns more
than functionality.

>
>>> A good source of NT reference material is Sean Daily's "Optimizing Windows
>>> NT" (IDG Books, ISBN 0-7645-3110-7): It's about $39 or so from the online
>>> book stores (amazon, etc...).
>>
>> Ah...the ever present third party book...someday, maybe MICROSOFT will be
>> the best source of NT documentation. Don't hold your breath. For a company
>> that wants to beat IBM in the enterprise, MS really has NO clue whatsoever.
>
> Novell is even worse. But in any case, do indeed thumb through this book
> at your local Borders or Barnes & Noble...
>
> Cheers!
> Dan
>
>
> -----------------------------------------------------------------
>
> THERE ARE NO ATTACHMENTS TO THIS MESSAGE, SO IF ONE
> APPEARS WITH IT, DO NOT OPEN OR DOWNLOAD IT!
>
> <mailto:expresso@snip.net>
>
> Webmaster for <http://www.Faulknerstudios.com>,
> <http://www.BrakeAndGo.com>
>
> This message is ©Copyright 1999 by Daniel L. Schwartz, and
> may not be reproduced except in its entirety.
>
> -----------------------------------------------------------------
>
> * Windows-MacOS Cooperation List *
>

* Windows-MacOS Cooperation List *

This archive was generated by hypermail 2.0b2 on Wed Oct 06 1999 - 07:42:31 PDT