Trade-Off Results for Connection Management

Abs t rac t . A connection management protocol establishes a connection between two hosts across a wide-area network to allow reliable message delivery. Following previous work of Kleinberg et al. (Proceedings of the 3rd Israel Symposium on the Theory of Computing and Systems, pp. 258-267, January 1995), we study the precise impact of the behavior of processors' clocks with respect to real time on the performance of connection management protocols, under common assumptions on the pattern of failures of the network and the host nodes. Two basic timing paradigms for docks are considered: clocks that exhibit certain kind of a drift from the rate of real time, and clocks that display a pattern of synchronization to real time; within each paradigm, several timing conditions on the clocks are assumed, giving rise to corresponding timing models. We consider networks that can duplicate and reorder messages, and nodes that can crash. We are interested in simultaneously optimizing the following performance parameters: the message delivery time, which is the time required to deliver a message, and the quiescence time, which is the time that elapses between periods of quiescence, in which a processor returns to an initial state and deletes all earlier connection records. We establish trade-offs between message delivery time and quiescence time, in the form of tight lower and upper bounds, for each combination of timing models and failure types. Severa] special cases of our trade-off results significantly improve upon or extend previous ones shown by Kleinberg et al.