| dc.description.abstract | This paper raises a fundamental questions, neglected so far in the literature: how to make a distributed algorithm robust against input errors and wrong probabilistic assumptions about the distribution of the inputs or of the link delays. We introduce a notion of complexity-preserving protocol controller: this is an automatic procedure that controls worst-case execution of any distributed algorithm. We then suggest a controlled with poly-logarithmic overhead. We show that the problem of designing controllers is a special case of another problem, referred to as dynamic resource management. We generalize our solution to solve the latter problem. We believe that the techniques used are basic ones, and will be used to solve a variety of unrelated network problems. Our solution seems to be very practical, since the formal code of the protocol is very simple and thus easy to implement. The technique used in the solution appears to be interesting because a global resource is manipulated locally. This somewhat resembles the "parallel prefix" technique used extensively in parallel computing. | en_US |
