File writes and long-running jobs

[Pat Wilson]

What strategies do you advise for users running long jobs and
writing output to AFS?  Much of the time long tokens are all my
"crunchers" need, but there's always the odd network glitch right
when they're writing their output (or the volume they're writing
to is being cloned, or the servers are restarting, or...).  What
do you tell them to do to mitigate these problems?

Pat