Cloudonomics

Weinman, an engineer and executive who worked at AT&T and HP, wrote a quantitative treatment of when cloud computing is economically preferable to owned infrastructure, using explicit models rather than assertion.

His central results concern demand variability. Owned capacity must be sized for peak demand, so it is idle in proportion to the gap between peak and average. Utility capacity is paid for only when used but carries a price premium per unit. Weinman derives the condition under which the utility is cheaper: it depends on the ratio of peak to average demand and on the utility premium, and he shows that pay-per-use wins whenever the premium is less than the peak-to-average ratio. He extends this to hybrid arrangements, showing that the cost-minimising configuration typically owns capacity for the base load and rents for the peak, and he derives the optimal split.

He then treats the value of on-demand capacity separately from its cost. If demand cannot be forecast accurately, both over-provisioning and under-provisioning impose costs, and he models the penalty function to show that the value of elasticity rises with forecast error. Other chapters cover the economics of distributed nodes and latency, the statistics of aggregating uncorrelated demands, parallelism and the value of completing work faster, and the reliability mathematics of redundant components.

The book is unusual in the cloud literature for actually doing the arithmetic, and the models are simple enough to apply. The limitation is that they capture cost and capacity while omitting most of what drives cloud adoption in practice: managed services, developer productivity, security and compliance, and the cost of operating infrastructure staff. The pricing environment has also changed substantially since 2012, and the repatriation debate about large workloads moving back to owned hardware has produced evidence Weinman's framework accommodates but does not anticipate.