Public Cloud Differential Pricing Design Under Provider and Tenants Joint Demand Response.

As the cloud computing market matures, effective pricing design remains a challenging problem. Electricity is a major contributor to the cloud's recurring costs and may increase significantly in the near future. As the cloud market gets more competitive, there will be a need for more cost-conscious dynamic pricing design. Additionally, today's cloud provider's pricing mechanisms act almost oblivious to the fact that the large tenants of the cloud are able to perform automated demand response in the form of delaying and dropping jobs. In this work we introduce differential dynamic pricing which provides different tenants with tailored prices according to their price sensitivity, while accounting for each tenant's energy costs. Furthermore, the cloud providers can engage in demand response to utility prices using on-site batteries. Battery management complicates things further, as it brings in additional costs, e.g., wearand-tear, which is challenging to attribute to each tenant. So, we use virtual battery management which accounts for individual tenant's usage of batteries. We also, show how virtual battery management can alleviate possible price fluctuations and act in synergy with differential dynamic pricing scheme and result in a less variable and more predictable (stable) pricing framework, called Stable Differential Dynamic Pricing (SD2P). We navigate different real-world IBM tenants' incident power demands with different price elasticities in conjunction with real-world electricity prices, and realistic battery degradation costs, to show the efficacy of our pricing mechanism. We show that under SD2P cloud's profit increases up to 31% while average tenants' utility increases up to 18%. Prices under SD2P exhibit up to 54% less variation compared to other dynamic pricing schemes, as a result of engaging the batteries at times of energy price spikes.