Framework for monitoring and testing web application scalability on the cloud
Proceedings of the WICSA/ECSA 2012 Companion Volume
Characterizing spot price dynamics in public cloud environments
Future Generation Computer Systems
Banking on decoupling: budget-driven sustainability for HPC applications on auction-based clouds
ACM SIGOPS Operating Systems Review
Deconstructing Amazon EC2 Spot Instance Pricing
ACM Transactions on Economics and Computation
Application-Centric resource provisioning for amazon EC2 spot instances
Euro-Par'13 Proceedings of the 19th international conference on Parallel Processing
| Hi-index | 0.00 |
In the Infrastructure-as-a-Service (IaaS) cloud computing market, spot instances refer to virtual servers that are rented via an auction. Spot instances allow IaaS providers to sell spare capacity while enabling IaaS users to acquire virtual servers at a lower price than the regular market price (also called "on demand" price). Users bid for spot instances at their chosen limit price. Based on the bids and the available capacity, the IaaS provider sets a clearing price. A bidder acquires their requested spot instances if their bid is above the clearing price. However, these spot instances may be terminated by the IaaS provider impromptu if the auction's clearing price goes above the user's limit price. In this context, this paper addresses the following question: Can spot instances be used to run paid web services while achieving performance and availability guaran-tees? The paper examines the problem faced by a Software-as-a-Service (SaaS) provider who rents spot instances from an IaaS provider and uses them to provide a web service on behalf of a paying customer. The SaaS provider incurs a monetary cost for renting computing resources from the IaaS provider, while charging its customer for executing web service transactions and paying penalties to the customer for failing to meet performance and availability objectives. To address this problem, the paper proposes a bidding scheme and server allocation policies designed to optimize the average revenue earned by the SaaS provider per time unit. Experimental results show that the proposed approach delivers higher revenue to the SaaS provider than an alternative approach where the SaaS provider runs the web service using "on demand" instances. The paper also shows that the server allocation policies seamlessly adapt to varying market conditions, traffic conditions, penalty levels and transaction fees.