Abstract
In some distributed systems, resources are leased, usually for a fixed period of time. For instance, a client leases a network printer for ten minutes. We consider the first step of leasing - acquisition, and extend the concept of the acquisition of a single resource to that of a collection of resources. In such a context, clients must have simultaneous access to all of the requested resources for the lease to be useful.This paper describes designs and implementations for collective acquisition of resources in distributed systems. It begins with the application background of our research, followed by the formalization of the problem. We then introduce our algorithm and prove its correctness. Two implementations are specified and compared. Evaluation of the performance of the algorithms is based on the measurements of the network overhead caused by the exchange of control messages, and the measurements of the average response time for the requests. Implemented in Java, our system makes novel use of multi-cast to enhance performance and uses heart-beat heuristics to achieve fault-resilience. Finally, we propose approaches to optimize the system performance exploiting soft global state information.