Problems with communication via PVM

[David L. Nicol]

Supposedly mosix load balancing plays perfectly nice with PVM.
A process that has been migrated under mosix has no idea that it is
not still running on the node where it was invoked.  For IO-intensive
jobs this sucks, because each system call has to go home to be handled,
but for CPU-intensive apps it works well.


After using PVM or MPI for initial process assignments, mosix can
act as a shock absorber, evening out the loading by transparently
migrating pvm processes to underloaded nodes.



Erik Arjan Hendriks wrote:
> 
> On Tue, Apr 25, 2000 at 09:38:47AM +0200, ala wrote:
> > Hi everyone,
> >
> >    I have huge problem:
> >      Have I written application (retracing) and I want paralleling it
> >      using bproc + PVM.
> >      I build PVM with BEOLIN settings and when it work without bproc
> >      all is correct, but when I move process to other computer or
> >      to this same machine, each sending data to the pvm generate
> >      strange unreadable message on the screen (some like "@ 43 _ ").
> >      I try use static and dynamic linked library, all nodes in cluster
> >      share this same tmp file.
> 
> PVM is not aware of bproc.  It doesn't expect processes to be moved
> somewhere after it's initialized.


__________________________________________________________________
                          David Nicol 816.235.1187 nicold@umkc.edu
                          "Rest is the sauce of labor" -- Plutarch
-------------------------------------------------------------------
To unsubscribe send a message body containing "unsubscribe"
to beowulf-bproc-request@beowulf.org