Pricing (and Bidding) Strategies for Delay Differentiated Cloud Services

We consider a cloud provider that seeks to maximize its revenue by offering services with different trade-offs between cost and timeliness of job completion. Spot instances and preemptible instances are examples of such services, with, in both cases, possible service interruptions delaying a job's completion. Our focus is on exploiting heterogeneity across jobs in terms of value and sensitivity to execution delay, with a joint distribution that determines their relationship across the user population. We characterize optimal (revenue maximizing) pricing strategies and, in the case of spot instances, optimal bidding strategies as well as identify conditions under which bidding at a fixed price is optimal. We show that correlation between delay sensitivity and job value needs to exceed a certain threshold for a service offering that differentiates based on speed of execution to be beneficial to the provider. We further assess the results' robustness under more general assumptions, and we offer guidelines for users and providers.