Capacity Reservation for Intermittent Random Demand Surges : A Model for Cost Optimization in Cloud Computing

Over the last decade, adoption of cloud computing has been accelerating, while firms are struggling to manage their growing cloud spending in the face of demand surges caused by planned or random events including marketing campaigns, new-product introduction, and natural disasters. In general, a firm can purchase capacity through a standard reservation contract from the cloud providers, supplemented by on-demand capacity as well as capacity purchased through reservation contracts from the reserved-instance marketplace (like the one administered by Amazon Web Services) to deal with the demand surges. We first analyze a model where the firm knows the distributions of the surge magnitude and duration. We characterize the firm’s optimal decision on when to sell excess capacity in the marketplace, show that the optimal capacity levels are dependent on the pricing of the reserved instances and the cancellation fee, and devise a heuristic for determining the length of the supplementary contract. Furthermore, we examine a model extension with the Bayesian updating paradigm, whereby the firm has a prior estimate about the surge-demand distribution, and we propose a pragmatic policy that leverages the firm’s trade options to adjust the desired capacity levels as data unveils and the estimate is updated. Through a comprehensive numerical study with simulation, we find that, to deal with the demand surges, the firm should purchase more capacity through a supplementary contract with a longer duration, if the unit discount rate increases, the cancellation fee decreases, or the average surge and inter-surge durations increase. In such cases, the firm also benefits more from the flexibility of trade in the marketplace, and the benefits achieve the maximum if demand stays at the base and surge levels for almost equal durations. Therefore, our study provides both efficient optimization algorithms and valuable insights into the firm’s cloud capacity management practices