| OCR Text |
Show 80 during I/O completion). Could this knowledge he taken care of? Solving this problem would be especially useful for future computer systems where the cost of arithmetic and control units is ex-pected to decrease much more than the cost of central memories. 3) Find models of "probable" performance as well as "worst case" models. 4) Which information other than a, 's or the b, .'s on the jobs would be relevant to an allocation algorithm? For instance, the exact virtual time at which a job will place a request might be available for some jobs whiüe bcinq completely out of the question for others. 5) How much would the results of the model be affected by slight errors in the predictions? III.7 Conclusion. My initial effort was applied to separate problems which are usually handled together in a very intricate manner: 1) Scheduling; 2) Paging algorithms; 3) Deciding external priorities of users 4) Col-lecting information about the average probable needs for resources of a specific job. Pricing, however, should not bo a question separated from scheduling. The problems of protection and of deadly embrace had already been separated from the others in previous work--:. Ry nartitionina the difficulty, I believe that the way to better sclontific understanding of shared computer systems stands open. The previous scheduling algorithms and models apnlv in computer systems where the shared facilities can either be preempted with very |
