Cost functions among five System/360 models are analyzed through examinations of instruction times, program kernels and a" typical" instruction mix. Comparisons are made between the data developed here and Grosch's Law which seems to be applicable to much of the data. Sizable economies of scale are unquestionably present in computing equipment.

:\most interesting question is that of centralization vs. decentralization in computing machinery. A survey of ('oinputer systems woul (l reveal a wide variety of configurations ranging from complete centralization in which onc la. rgc computer does most or all of the work for a firm to complete decentralization in which inany small com-1) utcrs fill the coinputation [leeds. Much disagreement exisls concerning this question. Some managers and specialists claim that highly centralized computing facilities are most econolnical while others claim the opposite. The purpose of tiffs paper is to shed additional light on the topic.

The subject of formal computer cost functions has received rather minimal attention in the past.'The reason is not that economists have been remiss, but that the problem has been complex. The output of computers is rather difficult to measure in any meaningful way'-'arid until recently each individual commercially available conlputcr was considerably different in operating characteristics, capabilities and configuration. In addition, comparisons anmng computers of different, manufacturers are exceedingly difficult because not only is the hardware different in kind but the other services that the manufactm'ers perform vary greatly in both quality and quantity. An increasingly important area of manufacturers' support is plvgramming effort. Some manufacturers supplS,