A short theory of multiprogramming

Pressures to use virtual memory are growing. Newer versions of application programs are growing explosively; for example, Microsoft Word, around 400K bytes in Version 3, is 25 megabytes in Version 6. Servers such as those running Windows NT use multiprogramming to share memory among jobs. We may see soon a revival of interest in the theory of multiprogramming as server administrators discover and worry about preventing thrashing. It is therefore useful to unearth the findings of the 1970s, which may be of interest to operating system designers and system administrators of the 1990s. How might one construct a self-regulating multiprogrammed virtual memory operating system that delivers near-optimal throughput? Such a system would necessarily not thrash. This is a hard problem. It is not obvious which design possibilities help solve the problem and are efficient to implement.<
>