Le jeudi 24 mars 2005 à 05:27 +0900, Adam C Powell IV a écrit : > Greetings and apologies for the long delay,
No problem, I know what it is. > > There is a patch available at http://squishy.monkeysoft.net/mpich/ to > > use mosrun instead of rsh. However, it isn't even necessary: mpirun > > should just understand --all-local and, when getting this option, spawn > > all processes locally instead of using rsh. OpenMosix then makes all > > processes migrate. > > There is a ch_shmem transport option in mpich which might suit your > needs. Please try building this option, by taking the mpich package and > editing debian/rules line 30, replacing: > --with-device=ch_p4 > with: > --with-device=ch_shmem > > If this works, then after the next bugfix release gets into testing, I > will upload a new version which builds an additional set of -shmem > packages to finally close this bug. But since that will go into the NEW > queue, it will probably take quite a while to get into the archive... I don't have an OpenMosix cluster handy to test that anymore, but I think this isn't suitable for OpenMosix, as processes that share memory can't be migrated. For best performance, I believe a backend using pipes or local sockets should be used. -- .''`. Josselin Mouette /\./\ : :' : [EMAIL PROTECTED] `. `' [EMAIL PROTECTED] `- Debian GNU/Linux -- The power of freedom
