Daemon's Advocate article

Dave Leimbach leimySPAM2k at mac.com
Tue Mar 2 06:33:53 PST 2004


Jonathon McKitrick <jcm at xxxxxxxxxxxxxxxxx> writes:

> Hey all,
> 
> the advocacy article at DNews had an interesting point: that the BSDs could
> end up being used only by developers, and not 'average' or should I say,
> typical users.
> 
> Do you agree or disagree with this?  Will DFly have any effect on the big
> picture?
> 
> jm
> -- 
> My other computer is your Windows box.

I think a lot of Linux/BSD is used and could be used by non-developers.

For my day job I do MPI implementations.  We more often than not have to deal
with researchers and scientists who just want the clusters to provide some
data for them and they go to some pretty far lengths to learn the one 
application they have to run beit NAMD for protein synthesis or what have
you.

If DragonFly, as a fork, helps them use a BSD in such an application better
than the others, then I'd say the job has been done.  I am not talking about
SSI [Single System Image] clusters either.  I haven't seen one of those in 
a research environment yet :).  And non-SSI clusters can range from sizes of
8 to 2500 or more boxes.  I know this first hand because I've helped install
both sizes.

My biggest fear is that in DraonFly's quest to become a great SSI platform that
it includes overhead that keeps it from being more loosely coupled in an 
efficient and meaningful way so that I can have my MPI cluster and people
who want SSI for some reason get their SSI :).  They are somewhat orthogonal
goals at some design level however things like checkpointing clearly have a 
purpose in both cluster layouts.

What is the most general purpose for an SSI cluster?  MPI clusters tend to not
be general purpose at all and really, in terms of application, more closely
resemble an embedded system where it is dedicated to one kind of computation.

Dave





More information about the Kernel mailing list