Software Controlled Access to Distributed Data Bases

AbstractControl of a distributed computing network requires an efficient method of allocating file resources, together with the capability of avoiding or recovering from deadlock. This paper describes alternative approaches to the deadlock problem, and examines two control schemes, one centralized and one distributed. Simulation shows that when perfect reliability is assumed, the centralized scheme is more efficient, except when traffic is primarily local in nature at each node. In practice, however, a centralized controller is often undesirable, since failure of the central node causes system paralysis. A modified control scheme is proposed, in which the centre of control can shift its location to adapt to failure of network elements. It is expected that such an adaptive controller will prove superior in performance to either of the previous alternatives.