Microsoft Releases Computational Cluster Technical Preview To olkit

Schilling, Richard RSchilling at affiliatedhealth.org
Wed Feb 14 13:28:11 PST 2001


Tim, 

We should also consider wether or not this is Microsoft's answer to the
clustering CD that was released at a trade show recently.  I think IBM
sponsored it.

see below

>Please try not to be so negative.  I was involved with producing the 
> "microsoft clustering solution", and I assure you, its much more 
> than you suggested.  

I know what I wrote sounds negative, but I've based my e-mail on a strong
technical and business opinion after a lot of continuous research,
performance analysis, etc . . .  I don't want to be flippant, and am glad
to speak openly about my concerns.  

>At the heart of the CD are the excellent tools from MPI software 
>Technology Inc.  They are commercial tools, so the Beowulf community 
>may not like them.  But rest assured, there are many companies out 
>there who'd rather pay someone to maintain their cluster middleware 
>than maintain it themselves.

I work for one such company.  It's common in health care to just pay
license fees, but it's also getting increasingly easier for a single
developer to develop the functionality for an entire industry.  That allows
a company to question the need to pay any license fee, even an $8.00 fee
for a CD.  I'm putting my $8.00 toward the purchase of Sun's Solaris free
operating system myself, and the clustering CD that IBM(?) has sponsored.

A great irony in this situation is that there are many excellent products
out there that run on Windows systems.  The problem I get into is when I
have to deal with the operating system itself, and Microsoft licenses.  But
that being said, the licensing fee MPI most likely paid to Microsoft to get
access to the information it needed to develop MPI for Windows will
increase the product's price.  I could be wrong on this, so tell me if I
am.  If MPI did pay a fee, I can do without the effective tax.  

> Also on the CD is Intel's MKL library, PLAPACK from Robert van de 
> Geijn's group, and a collection of scripts and documentation from 
> the Cornel Threory Center.  Oh, and lets not forget the Microsoft 
> contribution, Visual studio and Windows 2000.  

This begs the question - if I can get it elsewhere, or develop it myself,
why deal with the special handing Windows requires?  UNIX is the standard
here and common demoninator among these technologies, not the other way
around.  Why not use UNIX/Linux/FreeBSD?

Utilizing Visual Studio, dealing with installing/upgrading Windows 2000,
and working around any bugs is not something that will give me any
appreciable benefit for my current and future development projects.

An operating system that integrates web browsers, and other end-user
conveniences takes away from critical CPU time in a clustering application.
Visual studio is not compatible with the majority of GNU and UNIX
development tools, making it more cost effective to stay away from Visual
Studio.  And, when Windows incorporates other features the clustering
community (which includes Compaq/DEC, HP, IBM) have traditionally relied
upon on their UNIX platforms, I'll take notice.  

> Its a commercial clustering solution, so of course, the software is 
> provided as evaluation copies only.  But the evaluation period is 
> more than long enough to determine if the solution stack works.

But we cannot verify performance claims because the source code is closed,
making any technical claim dubious.  Again, it begs the question as to why
a company should spend scare capital on Microsoft software licenses -vs-
spending it on much needed employees and their training.  Being an early
adopter with Microsoft products, especially the operating system, is very
costly in terms of person hours, simply because the person adopting the
product has no choice but to work around the bugs.  That introduces extra
development time, etc . . .
Every product has bugs, but if I can't fix them it's a problem.

> Its important to note that there is a community out there --- 
> granted not well represented on this list --- who sincerely believe 
> in the Microsoft O.S.'s.   Rather than putting them down, I prefer 
> to recognize that "one size doesn't fit all" and do what can be > 
> done to enable clustering in both Linux and Microsoft O.S.'s.  

When the source code Microsoft provides is opened, both on the operating
system and the tools, then we can begin the serious work of enabling
clustering.  As long as I have to wait for the next version to come out
before I stop dealing with bugs, it's going to be very difficult to justify
a high performance solution to my employer.

On the business end of things, I know Microsoft is hoping the next version
of Windows, Windows XP, is going to boost new PC sales.  So, again comes
the question - why buy into a company that upgrades its products in a way
so as to require me to purchase a new computer?  The point here is to use
what we have as effectively as possible.  Also, if Microsoft has a product,
it also has an end-of-life in mind for the product.  The cost of migrating,
upgrading, in terms of licenses is just not financially beneficial to end
users.

--Richard



-----Original Message-----
From: Schilling, Richard [mailto:RSchilling at affiliatedhealth.org]
Sent: Tuesday, February 13, 2001 4:59 PM
To: beowulf at beowulf.org
Subject: RE: Microsoft Releases Computational Cluster Technical Preview To
olkit 


O.K. someone has to start, so might as well be me. . . . 

I'd bet if Peter Dell ran some numbers on how much it would cost to hire
all of us to develop custom software -vs- the total cost of ownership for a
"Microsoft clustering solution", we'd beat 'em in price.

I looked at the web page for the clustering preview kit, and the kit comes
with a bunch of evaluation software, including C++, and some versions of
MPI and PALPACK.  The problem I can see outright is that you've got to use
all the evaluation stuff - Visual C++, Windows 2000 Server, etc . . . to do
any testing.  And after all that work to develop a test suite (e.g. convert
your existing code), the only thing you have is a system that works for as
long as the evaluation software runs out - if you even get done before
then.  And buying the software to keep it going means buying into the
product life cycle for Microsoft products.  

I've seen too much work go into other products to justify the time needed
to explore this.  Perhaps just another marketing ploy to hock mediocre
software.  

So, am I alone in my concerns? 



Here's to getting Microsoft to CONTRIBUTE something USEFUL, 

Richard Schilling 
Webmaster / Web Integration Programmer 
http://www.affiliatedhealth.org 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.scyld.com/pipermail/beowulf/attachments/20010214/8278511e/attachment.html


More information about the Beowulf mailing list