A throughput model: Sequential vs concurrent processing

摘要

This paper is an account of experiments on information system architectures. Specifically, it deals with a comparison, under controlled conditions established by Cardenas, between the sequential and concurrent approaches to data-base manipulation. Cardenas' model was used to compare the throughput of sequentially manipulated databases. We here present a new model which permits the comparison of sequential and concurrent processing, using Cardenas' parameters and test data. The present model is devised in order to establish a means of validating the principle of hierarchical decomposition in the design of new information system architectures and to provide answers to three specific questions: 1.(1) What critical measures and parameters of information systems can be examined and controlled to optimize performance?2.(2) What attributes should an information system possess in order to achieve maximum throughput?3.(3) Under controlled conditions, what information system architecture achieves the higher level of through-put—a sequential (von Neumann) or a concurrent organization of processors?The results suggest a means of controlling response time, specifically in cases where the data-base has grown very quickly, and to maintain a level of response which can be proven to be cost effective. The results of these experiments have been used to validate a distributed architecture which forms the basis of the NDX-100 “electronic filing machine”.