Cloudonomics: The Business Value of Cloud Computing, + Website

Simultaneous Peaks

In the preceding discussion, we’ve assumed highly correlated demand. However, even when the demand is not highly correlated, there is a pathological case worth reviewing that has implications on both service provider economics and on customer agreements.

Consider two customers with varying demands D1(t) and D2(t) that are independent and uncorrelated: Their correlation coefficient is 0. Suppose, however, that for a single instant they simultaneously reach a peak. The correlation coefficient is still 0, but we now have a problem. If we are comparing a strategy of absolute assurance of sufficient resources via a build-to-peak strategy for each customer relative to a build-to-peak strategy for the aggregate demand for the service provider, there will be no net gain in utilization, and therefore no gain from statistical multiplexing. Although there still may be some net gain from economies of scale, as we’ve discussed, if the customers are both large enough to achieve substantial economies, there may not be enough of a further gain from consolidation to be either meaningful in and of itself or to overcome additional transaction costs, such as contracting for external services and the ongoing costs of service management and oversight.

This is a serious implication. If both customers must guarantee that sufficient resources are available and they have a simultaneous peak, regardless of what happens during 99.9999% of the time, that one instant can kill all the benefits ...