Abstract
This paper presents a method using which an asynchronous non-reactive distributed system can stabilize from a k-faulty configuration in a time that is a monotonically increasing function of k and independent of the size n of the system. In the proposed methodology, processes first measure the size of the faulty regions, and then use this information to schedule actions in such a way that the faulty regions progressively shrink, until they completely disappear. When k contiguous processes fail, the stabilization time is of O(k-cube). Otherwise, for small values of k, the stabilization time can be exponential in k, but it has an upper bound of O(n-cube). The added space complexity per process is O(d log n), where d is the maximum degree of a node.