The analysis and scheduling of devices having rotational delays

A number of problems concerning the scheduling, organization, and configuration of auxiliary storage units are analyzed in this dissertation. Stochastic, combinatorial, or simulation techniques are applied, depending on the assumptions and complexity of the particular problem. For the relatively simple scheduling disciplines of first-in-first-out (FIFO) and shortest-latency-time-first (SLTF), stochastic models are used. The starting addresses of I/0 requests to a file (non-paging) drum are modeled as random variables that are uniformly distributed about the circumference of the drum; the lengths of I/0 requests are modeled as random variables that are exponentially distributed. This model of I/0 requests is based upon measurements from an operational computer system. The arrival times of I/0 requests are first modeled as a Poisson process and then generalized to the case of a computer system with a finite degree of multiprogramming. Well-known resuits in queueing theory are sufficient for some models, but in other cases original approaches are required. In particular, a new model of the SLTF file drum is developed, is compared with previous models of the SLTF file drum as well as a simulation model, and is found to be a more accnreat model than previously available. Another practical problem that is discussed is an I/0 channel serving several, asynchronous paging drums. A new scheduling discipline is presented to minimize the total amount of rotational latency (and processing time) for an aribtrary set of N I/0 requests and the algorithm that is developed to implement this minimal-total-processing-time (MTPT) scheduling discipline has a computational complexity on the order of NlogN. The MTPT scheduling algorithm was implemented, and for more than three or four records, the most time-consuming step is the initial sorting of the records, a step also present in SLTF scheduling algorithms.