Keeping track of the latest gossip in shared memory systems

Abstract

In this paper we present a solution to the 'Latest Gossip Problem' for a shared memory distributed system. The Latest Gossip Problem is essentially one of bounded timestamping in which processes must locally keep track of the 'latest' information, direct or indirect, about all other processes. A solution to the Latest Gossip Problem is fundamental to the understanding of information flow in a distributed computation, and has applications to problems such as global state detection and mutual exclusion. Our solution is along the lines of that for message passing systems in [6], and for synchronously communicating systems [8]. Our algorithm uses a modified version of the consume and update protocols of Dwork and Waarts [3], where these were introduced to construct a 'Bounded Concurrent Timestamping System (BCTS)'. As applications of our Gossip Protocol, we also indicate another construction of a BCTS and a solution to the global state detection problem, which, we believe, are improvements over older solutions.