On designing direct dependency

Abstract

In this paper, we have proposed two recovery algorithms for distributed systems. Both algorithms follow a revolving centralized scheme. The direct dependency tracking of an integer representing the number of messages sent by each process has been shown to be sufficient to determine the maximum consistent state. The main feature of the recovery algorithms is that they are executed simultaneously by all the participating processes while determining the maximum consistent state. It thus ensures fast execution. The time overheads of the recovery algorithms are reduced further because both algorithms avoid some unnecessary comparisons while determining a consistent global checkpoint. The second algorithm has been shown to be faster than the first one, because it avoids, in general, much larger number of unnecessary comparisons compared to the first one; however the trade off is the increased amount of control information to be stored at each checkpoint in the second algorithm.