Long-term IaaS Cloud Service Selection

Abstract

There are two primary subscription models for IaaS cloud services: a) pay-as-you and b) reservation. Reservation-based subscriptions are typically offered for a long-term period such as 1 to 3 years. Long-term subscriptions are typically cost-efficient than short-term subscriptions for consumers who need services for a long-term period. Large organizations such as airline companies, banks, and research institutes tend to utilize IaaS services on a long-term basis for economic reasons. The performance of IaaS services is a key criterion to consider when selecting a service for a long-term. Selecting a service that may exhibit poor performance in the future may cause a significant loss of revenue for a business organization. Most IaaS providers, however, are reluctant to provide detailed information about their long-term service performance. This research aims at developing a long-term IaaS cloud service selection framework where IaaS providers reveal limited performance information about their services. First, we propose a novel framework to find the closest match of IaaS cloud service according to a consumer's long-term QoS requirements. The proposed framework leverages free short-term trials to discover the unknown QoS performance information. A temporal skyline-based filtering method is proposed to select candidate services for short-term trials. A novel cooperative long-term QoS prediction approach is introduced that utilizes past trial experiences of similar consumers using a workload replay technique. We propose a new trial workload generation model that estimates a provider's long-term performance in the absence of past trial experiences. The confidence of the prediction is measured based on the trial experience of the consumer. Next, we propose a new long-term IaaS cloud service selection framework that utilizes a consumer's trial experience and the performance fingerprints of IaaS cloud services for the long-term selection. We design a novel equivalence partitioning-based trial strategy to discover the unknown QoS performance variability of IaaS cloud services. A trial experience transformation method is proposed to estimate the long-term performance of an IaaS cloud service. Next, we introduce a signature-based IaaS cloud service selection framework that leverages a new significance-based trial scheme and a signature technique to discover a service's long-term performance. Next, we propose a novel event-based change detection approach to manage changes in IaaS performance signatures. A new anomaly-based event detection technique is proposed to detect changes in long-term IaaS performance behavior over time. We then propose an IaaS performance noise model to identify noise and actual changes in IaaS performance accurately. A novel categorical signature-based approach is proposed to detect the long-term performance changes using the proposed performance noise model. Finally, we introduce a signature change detection framework that leverages a sliding window-based approach and a Signal-to-Noise ratio-based approach to detect long-term changes in IaaS performance signatures. We have conducted a set of experiments based on real-world datasets to evaluate the proposed frameworks. The proposed long-term selection framework achieved almost 92% ranking accuracy. The signature-based IaaS cloud service selection framework achieved 96% ranking accuracy. The proposed changed detection frameworks achieved up to 90% change detection accuracy.