next up previous contents
Next: Input-output strategy Up: BASIC ARCHITECTURAL CONSIDERATIONS Previous: BASIC ARCHITECTURAL CONSIDERATIONS

Swapping

Our general picture of user programs included two types, the interactive ones and the noninteractive ones. The interactive programs would reside in ECS and from time to time be swapped into central memory and allowed to execute. (See Chapter 3, Hardware.) The noninteractive programs would spend most of their time on the disk. From time to time they would swap into ECS and while in ECS behave like interactive programs. If we were able to swap a program at the full ECS transfer rate of 10 words per microsecond, then a 10K word program could be swapped in and out in two milliseconds, allowing the use of a 20 millisecond quantum with 90 larger quantum to obtain the same efficiency, but even a 32K word program would need no more than a 64 millisecond quantum. Even if we assumed some degradation in efficiency of the swap, a maximum quantum size of 100 milliseconds would be sufficient. If there were 100 interactive programs, and 10 each would begin to receive service within one second. This seemed satisfactory.

Paul McJones
1998-06-22