SHARER, a time sharing system for the CDC 6600
Communications of the ACM
An appraisal of the Atlas supervisor
ACM '67 Proceedings of the 1967 22nd national conference
Exploiting the time-sharing environment
ACM '67 Proceedings of the 1967 22nd national conference
The SDC Time-Sharing System revisited
ACM '67 Proceedings of the 1967 22nd national conference
Microeconomics and the Market for Computer Services
ACM Computing Surveys (CSUR)
Introducing computing to smaller colleges and universities—a progress report
Communications of the ACM
A functioning computer network for higher education in North Carolina
AFIPS '72 (Fall, part II) Proceedings of the December 5-7, 1972, fall joint computer conference, part II
Some principles of time-sharing scheduler strategies
IBM Systems Journal
| Hi-index | 0.02 |
When a general-purpose computer performs compute-limited applications, its performance—unit cost per computation—is the primary measure of its adequacy. When such a computer performs a wide variety of services—interactive computation, fast-turnaround batch jobs, prolonged I/O-limited jobs, and prolonged compute-limited jobs—its raw performance is only one of several important measures: capability to overlap I/O operations with computation, accessibility to users, and turnaround times appropriate to user needs, as well as unit cost per computation. The focus of our paper is on two measures: capability to overlap I/O operations with computation, and a wide range of turnaround times. Consider the aggregate needs of a university population, representing the physical, behavioral, social, and engineering sciences, as well as the humanities and the university business Office. Some users—particularly in the behavioral sciences—will require the instant turnaround of interactive computing. Other users can be quite satisfied with ten-minute turnaround; they may prefer instant turnaround, but they can afford neither the terminal equipment nor the functional limitations imposed by interactive computing. Users with nontrivial jobs (at least three minutes on a powerful machine) will settle for turnaround of a few hours if—and only if—the unit price per computation is attractively low. Many administrative jobs and routine data-reduction jobs can be deferred until pre-dawn hours, i.e., over-night turnaround.