Record Details
Optimization techniques for time-shared computer systems
ScholarsArchive at Oregon State University
| Field | Value |
|---|---|
| Title | Optimization techniques for time-shared computer systems |
| Names |
Youssef, Mary Naguib, 1935-
(creator) Guthrie, Donald Jr (advisor) |
| Date Issued | 1970-03-12 (iso8601) |
| Note | Graduation date: 1970 |
| Abstract | The inefficiency of time-shared computer systems compared to batch processing systems is in the time lost in swapping operations. The larger the allocated quantum size, the less swap time is incurred. In order to guard against intolerable response time while lengthening the quantum size, the response time of a common request must be regulated. The criteria used in this paper to regulate the response time is to vary the quantum, with the number of users in the system, in such a way that the computer response time approaches the human response time. Based upon this concept, models are designed and analyzed to design an optimal scheduling algorithm which allocates the quantum dynamically. The models proposed are based upon Markovian assumptions for both arrival and service times. The priority discipline is round robin with dynamic quantum allocation. The swap time is assumed to be constant and the overhead time is zero. The inverse measure of performance is assumed to be the expected square difference between the cycle time and the mean human response time. In order to optimize these models two techniques are discussed. In the first, a mathematical optimization model is formulated in which a Markov chain is imbedded at the epochs of the beginning of a cycle. The cost function is assumed to be the inverse measure of performance. A technique suggested by Howard for optimizing a stochastic system under Markovian assumptions provides an optimal policy by which the scheduling algorithm allocates the quantum. The second technique discussed is based upon an optimal control system approach. The quantum size is chosen in such a way as to assure some stability property while improving system performance. A numerical example which illustrates these methods is provided. |
| Genre | Thesis/Dissertation |
| Topic | Time-sharing computer systems |
| Identifier | http://hdl.handle.net/1957/46036 |
