Cutting Your Cloud Computing Cost for Deadline-Constrained Batch Jobs

Many web service providers use commercial cloud computing infrastructures like Amazon for flexible and reliable service deployment. For these web service providers, the cost of cloud computing usage becomes a big part of their IT department cost. Facing the diverse pricing models including on-demand, reserved, and spot instance, it is difficult for web service providers to optimize their cost. This paper introduces a new cloud brokerage service to help web service providers to minimize their cloud computing cost for deadline-constrained batch jobs, which have been a significant workload in web services. Our cloud brokerage service associates each batch job with deadline, and always tries to use cheaper reserved instances for computation to maintain a minimum cost. We achieve this with the following two steps: (1) given a set of jobs' specifications, determine the scheduling of jobs, (2) given the scheduling and pricing options, find an optimal instance renting strategy. We prove that both problems in two steps are computation intractable, and propose approximation algorithms for them. Trace-based evaluation shows that our cloud brokerage service can reduce up to 57% of the cloud computing cost.