Wade Curry said:
> Hello,
>
> Still brainstorming, and coming up with ideas that may be wacky,
> but are sounding interesting to me at the moment.
>
> In recent history I saw an article in one of the Linux magazines
> about building a very small, cheap cluster. I'm thinking that I'd
> like to do something like this, but I'd have to work it in with
> my other project of making an application/file server.
>
> Anyway, beyond the fact that clusters share the workload between
> the nodes, I haven't really seen answers for some of my basic
> questions.
>
> 1) Does a cluster present a "single image"? By which I mean, do
> the nodes behave as separate computers that are just able to
> shuffle data and tasks back and forth? Or do they operate as a
> single machine in any practical sense?
Some may though I'm not familiar with any that work quite like that.
OpenMosix shuffles active programs around transparently, so I guess
you could call it a "single image" system.
> 2) There seems to be a tendency to have all the nodes identical. I
> believe this is just to simplify administration. Is this correct?
HPC clusters run bits of code in a parallel manner on slave nodes and
return the results back to a master node. Since most cluster software
uses some automated setup program, it's easier to use nearly identical
systems. Also it's easier keep spare parts around if all your nodes
are identical.
> 3) I understand there are several cluster solutions. I know about
> MPI, PVM, and OpenMosix. Is one better than the others? (in
> general, or for an application/file server)
MPI and PVM are not clusters. They are both essentially a set of
libraries that run on HPC (Beowulf) clusters.
MPI (Message Passing Interface)
A standard portable message-passing library developed in 1993 by a
group of parallel computer vendors, software writers, and
application scientists. Available to both Fortran and C programs
and available on a wide variety of parallel machines. Target
platform is a distributed memory system such as the SP.
PVM
A public domain message passing library called Parallel Virtual
Machine from Oak Ridge National Laboratory.
OpenMosix is completely different from HPC in that the program
migrates from one machine to another, seeking the least busy machine
to run on. It is tranparent to the user running the program. He/she
has no idea where the program is actually running, the results come
back to the workstation where it was initialized. There is no
master/slave relationship. It's often used to run complex programs on
a group high performance workstations, some which wouldn't be used to
their full potential, because they do simple tasks like word
processing. It speads up programs by adding unused computer resources
from other workstations.
They have different purposes so neither is inherantly better. I
suspect for most of us, OpenMosix would be more useful, unless you're
running a lot of programs that lend themselves to massivly paralles
processes. Most HP clusters are used for scientific calculations,
visualization, and other processes that can be parallelized.
> 4) Is it a good idea to have the workstations contribute to the
> cluster? OpenMosix seems to be good at that, from what I've heard.
> If I use Mosix on the wkstations, will it cooperate with PVM or MPI
> on the actual cluster?
>
I don't think that PVM or MPI even run on OpenMosix. They are
specifically designed for HPC.
Hope this helps.
--
Neil Schneider pacneil_at_linuxgeek_dot_net
http://www.paccomp.com
Key fingerprint = 67F0 E493 FCC0 0A8C 769B 8209 32D7 1DB1 8460 C47D
Sometimes I wonder whether the world is being run by smart people who
are putting us on, or by imbeciles who really mean it - Mark Twain
--
[email protected]
http://www.kernel-panic.org/cgi-bin/mailman/listinfo/kplug-list
