--- Paolo Giannozzi <giannozz at nest.sns.it> wrote: > > practical or unpractical, there is no other way to restart the phonon > > code. The problem is that it is impossible to distribute processes > across different processors in a predictable way, so the only safe > and simple solution is to have all files accessible from all > processes.
Actually, suppose the job is restarted on the same nodes as before, but the process ordering got screwed up. Would it be possible to do something like this: -upon a restart probe for the existence of the restartable files (the names are generally known!) on each node, and make a list of what is available -figure out if each process can see some unique file, and create a 1-to-1 mapping from process numbers to file numbers -then upon read/write simply remap the sequential file numbers into the other ordering Kostya __________________________________________________ Do You Yahoo!? Tired of spam? Yahoo! Mail has the best spam protection around http://mail.yahoo.com
