From: Josh Guffin Newsgroups: comp.parallel.mpi Subject: Re: [Q] Diskless execution with MPICH ? Date: Sun, 4 Jul 1999 22:40:27 -0500 Organization: Purdue University Message-Id: References: <377f5d7f.11406117@news.nic.utwente.nl> <377fdc1b.43809142@news.nic.utwente.nl> Mime-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII In-Reply-To: <377fdc1b.43809142@news.nic.utwente.nl> Xref: ukc comp.parallel.mpi:5282 I really haven't used PVM, so i couldn't comment on it's efficiency I also don't know if this would work, but if you're worried about disk space, perhaps just symbolic links to the appropriate executable would work... i.e. ln -s /mpiprog/myprogram /nfsmount/10.0.0.1/mpiprog/myprogram I assume you're using PC hardware and 10baseT or 10base2 ethernet, in which case i would assume that you should be fine. I'm certainly no expert, but IMO your network shouldn't run into any big bottleneck unless you're using a large (> 20 or so) number of diskless slaves. Good luck, Josh On Sun, 4 Jul 1999, Andrei A. Dergatchev wrote: > Thanks for your comments Josh, > > I've read today NFS and Beowulf HOWTOs, and the only > question left so far concerns my original MPICH/COW > combination - it seems to me that it will be inefficient > overhead to put on server several copies of the same > code for sake of MPICH and to spend time for its > transfer to clients by slow NFS ? > Can some one comment on this issue - since we have > relatively slow network cards and IDE HDDs, it is very > important for me to get other CPUs involved with > minimum losses (e.g. due to slow HDD of the server > and slow network I expect considerable slow down > for transfer of all /nfsmount/10.0.0.*/mpiprog) > Is PVM more efficient in that respect ? > > Thanks again, > > Andrei > =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= = Josh Guffin guffin@purdue.edu = = Purdue U. HEP - TASK E expert.cc.purdue.edu/~guffin = = #include = =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=