How to Bid the Cloud

Amazon’s Elastic Compute Cloud (EC2) uses auction-based spot pricing to sell spare capacity, allowing users to bid for cloud resources at a highly-reduced rate. Amazon sets this spot price dynamically and accepts user bids above this price. Jobs with lower bids (including those already running) are interrupted and must wait for a lower spot price before resuming. We answer two basic questions faced by providers and users: how will the provider set the spot prices, and what prices should users bid? Computing users’ bid strategies is particularly challenging: higher bid prices lessen the chance of interruptions, reducing the time overhead for recovering from interruptions, but can increase user costs. We address these questions and challenges in three steps: (1) modeling the cloud provider’s behavior to infer the spot price and matching the model to historically offered prices, (2) deriving optimal bid strategies for different job requirements and interruption overheads, and (3) adapting this strategy to MapReduce jobs with master and slave nodes having different interruption overheads. We run our strategies on EC2 for a variety of job sizes and instance types, finding that spot pricing reduces user costs by 90% with modest increases in completion time compared to on-demand instances.