R69-27 Demand Paging in Perspective

Suppose we have a computer system in which essentially all information is grouped for storage allocation purposes into fixed-length multiple word information units called pages. Suppose this system is being multiprogrammed or time-shared under the control of an appropriate supervisor system. For added flexibility and for implementing a so-called virtual memory (or one-level store), it is desirable to be able to execute programs without their being wholly in main memory. (Major examples of these systems include the Multics system of Project MAC and IBM´s TSS system.) The following general problem arises. Which pages of a program to be executed are to be loaded into main memory: a) just prior to execution, and b) at those subsequent instants during program execution when it is decided to bring in other pages because of references outside the initial set of pages? It is in this context that demand paging is defined and discussed by the authors whose paper is being reviewed here. (A related problem involves the decision as to which pages should be replaced, when this is necessary. Belady1 has made a detailed study of this latter problem.)