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

Vincent Diepeveen diep at xs4all.nl
Mon Jun 12 10:14:19 PDT 2006


I'm under the impression you're not quite understanding what microsoft has 
been doing.

They've put in a few billion into server software development and plan to 
take over the
entire server market.

The cluster edition is just 'part' of their plan to do that.

How that will scale and run from performance viewpoint, that's just not 
interesting
from users viewpoint.

The real important thing is that most new generation people know how to
work with a mouse and click 'ok' in a dialog box in windows.

That's why of course all that software will move to windows.

The alternative, some textmode app in linux which no one can work with 
except
a few geeks on this list, will simply not get used in future.

It will all move to microsoft.

And microsofts 'grand master plan' to take over entire server market is not 
something
you must see in the short term, but in the long term.

In the long term they will scale better perhaps after doing 5 bugfixes here 
and there and
shipping messages in windows to other processes is quite easy.

There is 2 main API commands for that is for example:

  sprintf(buf,"Loading engine\n");
  PostMessage(hwnd,ENGINE_TEXT,0,(LPARAM)buf);

So microsoft can do a few bugfixes and get it to work a bit faster that 
scales a bit better then.

I already had to choose whether to develop a GUI that works only for windows 
or a portable GUI that works on
both. Though our GUI can be ported to linux, i can't sell enough products 
simply for linux to pay for the GUI
porting cost, let alone support.

So we dropped directly that idea.

This is why windows will dominate simply. User friendlyness. When that means 
big problems for us as developers
and seeing all the bugs in windows, that's not real interesting for the 
users. User

If their choice is buying a linux ferrari in 1000 parts, under the condition 
that before each time they go drive with their
ferrari they know how to put together those 1000 parts, and as a second 
choice a windows toyota camry that works
under all circumstances, then the average user will opt for windows.

If another disadvantage for an user is that he has to reboot each day his 
cluster when it runs windows, that's not a real
problem for the user, as he's used to turn off his computer at the end of 
each day anyway, so he'll turn off the cluster
each day anyway and boot it in the morning.

But in the long term it is the user who decides what he is going to use. Not 
you nor me.
If your 2 options are to buy a windows edition and windows commercial 
software or make software yourself,
what will YOU do?

Vincent
----- Original Message ----- 
From: "Chris Dagdigian" <dag at sonsorol.org>
To: <beowulf at beowulf.org>
Sent: Monday, June 12, 2006 4:03 PM
Subject: Re: [Beowulf] MS HPC... Oh dear...


>
> As usual, RGB nails it.
>
> Anyone reading this list is not a candidate for MS Cluster Server  2003 - 
> the main target is commercial software vendors and possibly  some 
> specialized turnkey system integrators.
>
> One of the markets I can sorta see for MS in my field (life science- ish 
> stuff) is the market for CPUs that do pretty much nothing but  service the 
> needs of exotic lab instruments.
>
> There is a pretty good mix of six and seven-figure lab hardware that  in 
> the past used to ship with a dedicated workstation to handle data 
> collection and first/second-pass processing of results. These systems  (in 
> some cases) are now shipping with small clusters to handle the  processing 
> demands. The people purchasing these lab instruments are  straight wet lab 
> biologists or chemists and they really could care  less about the data 
> collector/processing box.  They've already spent  north of half a million 
> dollars just to get the instrument and its  related infrastructure and 
> they really do not want to be in the  business of rolling and managing 
> their own lab cluster. They want to  run their instrument, not screw 
> around with its related infrastructure.
>
> Take one of those nice looking Rocketcalc deskside cluster boxes or 
> something like that Tyan cluster chassis that was discussed on the  list 
> last week,  slap MS Cluster Server on it and pair it with a  nicely 
> supported (by the instrument maker or some third party)  software stack 
> that supports and drives the exotic lab instrument and  (I think) people 
> will buy them.
>
> The interesting thing for me is at what level MS will consider this a 
> success or failure.  Sales of 20,000 licenses in a year may translate 
> into "horrible failure to penetrate the cluster market" within MS but  it 
> may represent an absolute windfall to small and specialized  software 
> shops who only need to sell a few dozen or a few hundred  licenses to 
> their very specialized market niches per year. For the  small ISVs, even a 
> MS "failure" could be very very significant in  terms of volume and 
> revenue. To them, MS Cluster Server 2003 is  nothing more than a delivery 
> platform for their niche codes,  potentially one that is easier to build a 
> sales and support model  around than a pure Linux cluster would. Their end 
> users / target  market would not really care about the base OS.
>
> my $.02
>
> -Chris
>
>
>
>
> On Jun 12, 2006, at 10:30 AM, Robert G. Brown wrote:
>
>>
>> So, naaaaa, not likely to be a popular development platform for real
>> researcher's writing their own code or using open source code.
>> Commercial only.
>
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or unsubscribe) visit 
> http://www.beowulf.org/mailman/listinfo/beowulf
> 




More information about the Beowulf mailing list