FORECASTING DEMAND FOR CLOUD COMPUTING RESOURCES - An Agent-based Simulation of a Two Tiered Approach

As cloud computing grows in popularity and usage, providers of cloud services are facing challenges of scale and complexity; how can they ensure they are most efficiently using their existing infrastructure, and when should they invest in new infrastructure to meet demand? We propose a two-period model which utilises a third party called the Coordinator, who interacts with a population of resource-buyers. The Coordinator uses two mechanisms to aid the provider in future capacity planning. Firstly, the Coordinator extracts probabilities from the buyers through an options market to determine their likely usage in the next period, which can subsequently be used to schedule workloads. Secondly, the Coordinator uses previous market demand to predict if cost can be reduced by investing in a reservation over a longer period. This upfront investment contributes to the provider‟s capital expenditure in new capability and implies that Coordinator intends to further utilise such an investment. We implement the model in an agent-based simulation using actual UK market data where a pool of users submit different probabilities based on previous market demand. We show that the Coordinator can make a profit when faced with different market conditions, and that profit can be maximised by considering the utilisation of previously purchased