IMPLEMENTING A LARGE VIRTUAL MEMORY IN A DISTRIBUTED COMPUTING SYSTEM

Local area networks are usually designed as extensions of the conventional idea that long-lived objects (i. e. files) are treated separately from the computational short term objects of executing programs. We propose an alternative technique which supports a single very large virtual space with unique addresses across a network. The unit of information transfer between nodes is a fixed length page and the design is so organised that a range of configurations can be supported without modification to the software or hardware. At one extreme, a single node might act as the sole 'memory' server (analogous to a file server), while at the other extreme each node might 'own' some part of the virtual memory and make it available to other nodes on request. To upgrade an operating system designed to work stand-alone with a unique virtual memory requires only changes to the paging software and the process scheduler. Higher level software (e. g. directory modules, command language interpreters) remain unaffected