A feature of fault tolerance can be incorporated in the farm daemon. This can
be used when a node having a worker task goes down. PVM provides constructs
to notify a task (if it has so indicated) when a task exits PVM or the node
on which it is running halts due to some error [7] . This feature could be useful
in the case where the farmer task needs ``at least one'' reply to come from
the worker task for every work packet that is sent out. It would be possible
to incorporate fault tolerance in the farm daemon because it could maintain a list of work
packets which were currently being processed (it needn't keep a list of all
work packets, as when a worker task requests for a new work packet, the earlier
one can be removed from the list). When PVM would notify the farm daemon of a worker
task exiting or a node malfunctioning, then the farm daemon could either fork the
task on another node or using the existing nodes could resend the work packet
to the other worker tasks after removing the entry about the dead task(s) from
its lists.