Section: New Results

Idleness and failure prediction in large infrastructures

We have proposed a method to discover statistical models of availability in large distributed systems and applied it to run an enlightening study of SETI@home [19] . This was also used to make long-term availability predictions for groups of desktop grid resources [21] . We have used statistically based models of heterogeneous failures in parallel systems and assessed their tolerance [39] . A similar approach was used to design correlated resource models of Internet end hosts [38] , [17] .