From: Clark Dorman Newsgroups: comp.parallel.pvm Subject: Re: Fault tolerance Date: 30 Dec 1998 08:16:49 -0500 Organization: PSINet Message-Id: References: <76b019$1d7$1@news.flashnet.it> Cc: f.piantone@flashnet.it "Faber" writes: > I'd like to know more infos about: > Fault tolerance in a distributed system under PVM and Linux386 . > Reference and other. I cannot give you a reference because I do not have one. In my experience though, the PVM system can be fault tolerant for every system except the one that has the master daemon. If any other machine or process in the parallel machine dies, it is possible to have other processes eventually notice this and error recover and re-start the process. My distributed system does this, periodically listening for messages that basically say "I'm still alive". If the main program does not get such a message, it will go and look for the process. If the process does not exist, or the machine's daemon is gone, it will start over. Note that you have to do this yourself. PVM is a low-level system and while it gives you great flexibility, it also requires you to do things, like check for faults, yourself. For the main daemon, however, the system is not fault tolerant at all. If the main daemon dies, or if the system that has the main daemon crashes, the rest of the daemons and processes are lost. I do not know how to solve this problem, because it seems to be inherent in the design of the system. I hope that Harness will have a system organization that will correct for this. -- Clark