A reexamination: Centralized versus decentralized computers in biomedical environments

Abstract

We feel that, rather than asking whether biomedical investigators should be supported by centralized or decentralized computers, we should ask how investigators may be given effective access to the various kinds of computing service they require. Optimality of a computer configuration cannot be meaningfully assessed apart from the job-mix priorities it is to serve. It is difficult to think of an area of application wherein requirements would differ more from location to location than in biomedical computing. What is good for control of on-line experiments in a physiological laboratory would tend to be inefficient or altogether inadequate for processing the major statistical computations widely encountered in biology and medicine. Conversely, it would usually be uneconomical to tie up a system well suited to statistical computations for continual monitoring of experiments. Not only do we find a spectrum of biologists whose requirements range from control of experiments to complex modeling and statistical analysis, we also find the entire spectrum represented in the work of a single investigator—often, in fact, in a single study. Thus, after preliminary reduction by the laboratory computer, the neurophysiologist's data may require major statistical analysis. In fact, it is likely that occasional bursts of such major support during the course of an experiment might be desired for more sophisticated guidance of the experiment.