On Sunday 02 December 2001 15:01, you wrote:
> Yes and no.
I'll add to this. Our 16 nodes cluster is a 3-tiered approach.
LAM/MPI for the big stuff
PBS (queueing system) for most 'small' jobs (for 5-25 hours/cpu values of
small ;)
and MOSIX for whatever else, especially interactive usage of the master
console.
MOSIX cannot handle sockets and parallel jobs too well. I mainly use it as a
last-resort measure, anything PBS or LAM starts I modified to tell MOSIX to
look the other way.
MOSIX is nice for interactive usage, on big private networks (like say a
commercial scientific research lab -- MOSIX the workstations together) but I
wouldn't use it as a basis for developing large-scale applications like
weather models and such. Not when I have MPI and queueing systems handy
anyway.
--
Christian Lavoie
clavoie_at_[hidden]
_______________________________________________
This list is archived at http://www.lam-mpi.org/MailArchives/lam/
|
