Enterprise computing: economies and diseconomies of scale in server price/performance relationships

It has been long held that computer systems exhibit economies of scale (Grosch's law) and that their price increases with performance, but at a decreasing rate. The empirical support for this proposition has largely been based on benchmark tests of complex processor performance engaged in a sequence of mathematical computations. This paper estimates price/performance elasticities using performance data derived from benchmark testing that simulates on-line transaction processing of the type found in business environments and analyzes the data by series, rather than by arbitrary hardware classes. We find that the systems of at least two series have price/performance relationships characterized by economies of scale, constant returns to scale, and diseconomies of scale, respectively. That is, Grosch's law holds for some series and not for others. Second, there is a direct relationship between performance and the magnitude of elasticities, with high-end servers exhibiting the largest percentage increases in price for a given percentage increase in performance. A discussion of other life cycle cost and performance issues is provided to give explanation and insight into the market place presence and investment in systems that possess diseconomies of scale based on price (initial cost).