[Beowulf] MS HPC... Oh dear...

[Joe Landman]

Gerry Creager N5JXS wrote:
> Joe, et al,
> 
> Unfortunately, we find, time and time again, when Microsoft comes in to
> tell us how good they're gonna do something, there's a catch.

As with most companies, there is.  They are trying to sell you something. :(

> Here, it remains simple to me: I can tweak the OS if I need to, and I
> can tweak the apps.  With a Windows solution fo rmy cluster environment,
> I certainly can't tweak the OS.  I'm half-way to dead in the water.

I agree with the idea of tweaking the OS.  We do lots of tuning for our
customers.  Usually it starts out with the right kernel and config and
works from there.

> In general, we integrate our own hardware.  If I found the API was too
> complex, I'd complain mightily, but actually, I tend to pick hardware I
> can already talk to, and API's I'm already familiar with to program to.

To a degree this is my point.  Microsoft (I am not arguing their case,
just my impression of it) is going to try to make all this work out of
the box for you.  It remains to be seen how well it works.  I can't wait
for the first support calls about the ch_p3/p4 device though ....

>  At this point, Microsoft is claimint to save me from myself.  Hmmm. I'm

Don't know about that, can't/won't speak to that.  The argument for them
is a "simpler" (monocultural) worldview.  The argument against them
includes the monoculture :) (marketing ju-jitsu), and that most of the
new development is in Linux, and Linux clusters have been growing
rapidly and driving this market to the heights it is at.

> reminded of the time IBM decided... and advertised... their intent to
> "legitimize" the minicomputer industry.  If I recall correctly, the Data
> General ad in response included the comment, "The bastards say,
> 'Welcome'."  That's sort of what I feel here.

Heh.  Might be worth a nice banner advert somewhere.

> I'm reminded that Microsoft made attempts to wander into this community
> several years ago.  I further recall being simply told that there was no
> direction for Linux, no core of paid developers, and that the HPC work I

I remember that some booth-droid was telling me this almost 3 years ago
at SC03.  I asked them simple questions about costs, about performance,
and all I got was hemorrhagic marketing.

I appreciated speaking with the folks I did last week.  I didn't agree
with all they said, but it is clear they are at least looking at the
problems.

> was doing couldn't expect to continue using that paradigm. Simply, I
> should migrate my applications to Windows (NT4, at that time) and wait
> for  the Microsoft HPC tool.

Like Microsoft F90/F95 ... :)

> 
> I'm not sure I'm ready for Microsoft to define my timeline and my code
> requirements.
> 
> I'll stand by my position.  The bulk of the Microsoft announcement is
> marketing, 

Yes.  Agreed.  All announcements are this.

> and demonstrates a less than stellar acknowledgment of the
> HPC environment they're trying to break into.  If y'all think your
> corporate wagon's hitched to their star, that's great.  I'll watch and

Nope.  No hitching going on.  We follow what our customers ask for.
Thus far it has been fast linux clusters.

> learn.  I'm not completely wedded to Linux, but I'mnot likely to change
> until it does something similar to a series of data losses eventually
> identified as registry corruption which encouraged me to make the swap
> to Linux.  As long as the problems I have continue to be hardware
> related (think, statistical failure rates) and not OS definable, I doubt
> I'll seriously investigate them.

Yes. Its not the OS.  Its the stuff atop the OS that matters.  Its how
well it fits.  Its how well it works.
Not having seen WCC, I don't know what its like.  I would be hard
pressed to imagine that the environment is similar to Linux in any
significant way.  That is, I cannot just take a source code with gnu
configure, run ./configure (or its windows equivalent) , and build
things.  So porting over to it is non-trivial.

If they want adopters, they need to make transitions painless.  I am not
sure it will be painless.


-- 
Joseph Landman, Ph.D
Founder and CEO
Scalable Informatics LLC,
email: landman at scalableinformatics.com
web  : http://www.scalableinformatics.com
phone: +1 734 786 8423
fax  : +1 734 786 8452 or +1 866 888 3112
cell : +1 734 612 4615