[Beowulf] Joe Blaylock's notes on running a MacOS cluster, Nov. 2007

Geoff geoff at galitz.org
Wed Nov 21 07:51:10 PST 2007

> One day Apple will make up its mind about whether it is a hardware
> company or a software company.  I vote soft, and thought that once they
> supported generic Intel even their main company management would "get
> it".  It continues to think "hard" though, probably because they
> armtwist a significant premium out of their customers for buying their
> stuff and because it locks many of those customers mentally into buying
> apple forever as they identify apple AS a computer company and say they
> like apple computers, when what they really like is apple's software.

I have to disagree.  If you are talking about Apple surviving, they have  
proven that their current approach works very well.  If you are talking  
about their penetration into HPC (where I have in fact seen a number of  
Mac/Apple clusters) then their current approach just does not target HPC  
effectively... nor should it.  Being all things to all people (or markets)  
results in too many compromises of design and execution.

While their OS is indeed constrained to their hardware, you can freely  
install and run other OSs on their hardware (Windows, Linux and *BSD) so I  
would agree they tend to position themselves as a hardware company but I  
disagree that they will need to make a choice in the future about that...  
unless market conditions change, of course.

I think more companies and vendors need to better focus on the integration  
between hardware and software.  Being one or the other results in too many  
gaps in the implementation of useful technologies.  Having a default  
choice with a good track record and a clear focus is a "really good thing"  
(TM) in my opinion.  It does increase the cost of the computing platform  
but there is plenty of room in the universe for that.

> Sun needs to make the same choice.

Perhaps.  In my view they have made some unfortunate compromises.  The  
lack of floating point performance when the Niagra processors hit the  
market was a bad compromise, for example. I am sure they saw their AMD  
offerings as a balance to that. Their initial implementation of a cheap  
3rd party IPMI/SMDC solution for their x2100 series systems has been a  
huge PITA for my current client.  There is such a thing as too much choice  
and Sun was (and perhaps still is) offering too much for customers to be  
able to make an informed purchasing decision.


Geoff Galitz, geoff at galitz.org
Blankenheim, Deutschland

More information about the Beowulf mailing list