Exploring the profit-reliability trade-off in Amazon´s spot instance market: A better pricing mechanism

In Amazon´s spot instance (SI) market, the volatility of the two important parameters, namely spot price (SP) and inter-price time (IPT), affects not only the market´s profit, but also its service reliability. Thus, it is important for the cloud service provider to understand how SP and IPT impact the profit-reliability trade-off in the SI market. To the best of our knowledge, such a trade-off has not been studied in existing liter-ature. In this paper, we model Amazon´s SI market as a modified repeated single-price auction and study the profit maximization problem using a graph model. We prove the NP-Completeness of the corresponding offline decision problem. Moreover, we propose an order-statistics based online pricing (OSOP) algorithm that can effectively evaluate and tune the profit-reliability trade-off by on-the-fly adapting SP and IPT. In our approach, SP and IPT are determined in real time based on the order statistics of the latest historical bids and the profit-reliability trade-off desired by the service providers. Our experiments show that the proposed OSOP mechanism (on average) achieves as high as ≈19% profit gain as compared to the current algorithm, with negligible reliability loss. Moreover, the mechanism also achieves a favorable trade-off between profit and service reliability, at which point our mechanism (on average) achieves ≈ 12% profit gain and ≈ 8% reduction in unexpected service interruption penalty as compared to the current algorithm.