philosophical question - limits of beowulf defn

[Schilling, Richard]

Dr. Ligon presents great points. . . 

Great discussion, and I'd like to interject a little $.02 from the health
care industry.  
The attractiveness of beowulf technology lies in its definition: beowulf
refers to a 'class'
of machines, and general topology, which I believe is what the folks at NASA
intended.

That's important, because it lets me adopt the approach to what works best
for my environment.
Case in point, PVM, MPICH, JPVM or even Serialization of Java classes would,
as I understand 
it be appropriately used in a beowulf cluster.  That flexibility allows 
me to run proof of concepts on the hardware I have available.  

The general definition of beowulf-class machines also lets me focus on the
functionality of 
my application more than the particular toolset used.  Speed of prototyping
is crucial to my research 
efforts where clustering in and of itself is not my core objective.  So, a
general purpose
definition rather than a detailed specification is important - specific
benchmarks, although 
an initial consieration, can be improved later.

At the same time having a standard reference to a beowulf-class machine
gives me an 
indespensible frame of reference - and one I'll desperately need once I
present my initial results
to the boss to ask for more hardware!!


Richard Schilling
Web Integration Programmer
Affiliated Health Services
Mount Vernon, WA  USA


-----Original Message-----
From: Walter B. Ligon III [mailto:walt at parl.ces.clemson.edu]
Sent: Monday, July 17, 2000 10:56 AM
To: David S. Greenberg
Cc: Ward William E PHDN; 'Beowulf' Mailing List (E-mail)
Subject: Re: philosophical question - limits of beowulf defn 


--------
A quick agreement on both Doug's contribution and David's posting.
The EL community did grow up as a SUPERSET of Beowulf because there
are many cluster based systems that are not, technically, a Beouwlf,
but share many interesting problems.  I think it is even MORE important
to realize that the EL community is ALSO focused on the open-source (Linux)
OS issue - which allows us to evolve the OS to better meet the needs of
more extreme systems (one of which is a Beowulf).

I have always been rather particular about the name "Beowulf" for two
reasons:

first, people did cluster computing long before Beowulf.  When Beowulf
	came out many people said "so what? - it's been done before."
	but they missed an important point in the COTS HW and open-source
	OS!  The specific characteristis of a Beowulf mean we are *free*
	to adapt the architecture because the nodes are dedicated, because
	the OS is open-source, because parallel computing is our goal.  I
	think the next-gen Beowulf software we are working on is going to
	drive that point home like never before.

second, there was an early tendency for people to say "everything is a
	Beowulf."  I'm not against everyone joining the party, by there are
	many DIFFERENT approaches, including high availability designs,
	true distributed systems, cycle stealing, etc. and they have
	different issues.  I don't want to dillute the effort going into
	ANY of these areas byt saying "Hey, its ALL Beowulf, so we're doing
	Beowulf and we don't need to do any more."  That's not where its at.
	Beowulf is in its infancy, and it is NOT the answer to everyone's
	problem.  I would rather ORGANIZE our efforts so that ALL of those
	efforts get the recoginition they deserve than let it all wind up
	a big cloudy mess!

So that's why I always butt in on these discussions and why I always take
the position that I do.  The folks that defined the term had a specific
thing in mind, and they have repeatedly indicated they want that definition
to stay true (even though some of them are now off doing DIFFERENT things).
Some day they may come down and tell me to shut up - but until then, I will
try to let everyone know where this term came from.  I'm not trying to be an
ass-hole, I just want to let you know.

Walt

-- 
Dr. Walter B. Ligon III
Associate Professor
ECE Department
Clemson University



_______________________________________________
Beowulf mailing list
Beowulf at beowulf.org
http://www.beowulf.org/mailman/listinfo/beowulf
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.beowulf.org/pipermail/beowulf/attachments/20000717/ceb8b2ec/attachment.html>