Understanding Non-Uniform Failure Models
Petr Kuznetsov · Bulletin of the European Association for Theoretical Computer Science · 2012
Traditionally, models of fault-tolerant distributed computing assume that failures are “uniform”: processes are equally probable to fail and a failure of one process does not affect reliability of the others. In real systems, however, processes may not be equally reliable. Moreover, failures may be correlated because of software or hardware features shared by subsets of processes. In this paper, we survey recent results addressing the question of what can and what cannot be computed in systems with non-identical and non-independent failures