An Architecture for a Local Network Providing Multiple Supercomputer Access

Abstract An architecture for a non-homogeneous local network cormecting a large number of users(workstations) to a numberofservers (some of them supercomputers) is described in this paper.A connecdonless communication model is assumed and a high speed broadcast channel is used asthe imerconnection network among the servers and the users of the network. The AvailabilityDriven Multiple Access Architecture is based upon the correlation of the channel allocationstrategy with server availability, by means of scheduling protocols. A scheduling protocoldefines a communication discipline in which servers capable ofproviding services, compete forconttol of the communication channel by using a multiple access algorithm in one of thefollowing classes: random, limited contention or contention free multi-access.When in controlofthe channel, a serverperforms a sequence ofdifferent activities, including a broadcastmessageofits willingness to provide certain types ofservices, thus inviting all users to send their servicerequests. Then, a multi-accessalgoriilim is used to give each user a chance to send its request tothe remote server.