Sören Asmussen, Aarhus University, Denmark

Limit theorems for failure recovery probabilities

A job like the execution of a computer program or the transmission
of a file on a communications link may fail. Some of the schemes
for failure recovery that have been suggested go under the names
of REPLACE, RESUME, RESTART and checkpointing. I will start by a 
survey of a recent analysis of RESTART (joint with  Fiorini, 
Lipsky,  Rolski & Sheahan), which shows that this scheme may lead 
to total completion times with a disasterously heavy tail. One way 
to mitigate this is checkpointing, where the job is split into subtask 
so that a it suffices to restart from the last checkpoint. We investigate
the effect of this on the tail of the total time distribution for a 
variety of checkpointing models (joint with Lipsky).