Robust coordination in distributed multi-server systems

One way of substantially improving the reliability and performance of parallel and distributed applications is to replicate a server over multiple machines. Examples of multiple server systems include operating system support for parallel machines, distributed database systems, and multi-agent robotic and manufacturing systems. Coordination is required for parallel access to shared resources such as data structures, communication network, and physical space. Coordination is also required to ensure adequate speed-up when multiple servers are harnessed to work on different portions of the same task. The authors present two coordination algorithms for task distribution among identical servers, namely, a ticket-based algorithm and a toroidal graph algorithm. These two algorithms are highly robust and efficient. They allow uniform task distribution over working servers even in the presence of up to N - 1 server failures, where N is the number of servers. The ticket-based algorithm has a time complexity O(log N) and the toroidal graph algorithm requires only constant time.