A cost analysis of multiprogramming and swap-time in a time-sharing system

Abstract

In a t~me-sharing system the most important aspect of performance is the response time to the user. The question of making improvements to the hardware must be evaluated in terms of how it will improve the response time to the user. The average cost of a single interaction (C I) for N active term%nals including the cost of the user's time at th~ terminal is: where CII = CS* (R+T k}/T+N*H* (R+T k)/T C S = Hardware cost for t~me period T W = Wages of user for t~me period T R = Average response time and T k = average think time R+T k = Average time for one interaction assuming additional hardware cost is ~&C S cause a decrease in the response time AR, then C12 = (Cs+ZICs)* (R-LLR+Tk)/T+N*W* (R-~+Tk)/T if ~C S <~R/CR-~R+Tk)*(Cs+N*W) then C12 < CII The above relation shows that an additional hardware cost~C is cost effective if it is less th~n AR/(R-AR+T k) times the cost of the original system plus the wages of N persons. The cost may be lease or purchase over any time period as long as the same time period is used for Cs,~C S and W. Figure I. shows the flow of a process associated with a specific terminal in a time-sharing system. Once a user completes his thinking at a terminal he responds to the system by typing input at his k~o His request for service waits on the swap queue until the code associated with his process can be read into memory. The request waits on the swap queue until all requests in front are serviced. Upon the completion of a swap operation the process must wait on the CPU queue for the use of the CPU. Once the process is given control of the CPU it willterminate by requesting input from the terminal. All queuing is first come first serve. The time for an interaction is the time to complete a full cycle in Figure I° The response time is an interaction time, minus the think time or.the swap queue time, plus the swap t_ime, plus the CPUqueue time, plus CPU time.