Component failure in large-scale IT installations such as cluster supercomputers or internet service providers is becoming an ever larger problem as the number of processors, memory chips and disks in a single cluster approaches a million. In this paper, we present and analyze field-gathered...