IaaS reserved contract procurement optimisation with load prediction
Faculty of Sciences. Mathematics and Computer Science
COMPUTING AND ESCIENCE
Future generation computer systems: the international journal of grid computing: theory, methods & applications. - Amsterdam
, p. 13-24
University of Antwerp
The increased adoption of cloud computing, combined with the recent proliferation of pricing plans has increased the relevance of automating the complex and time consuming tasks of selecting, procuring and managing cloud resources. In this work, we present an approach to automate the procurement decision of reserved contracts in the context of Infrastructure-as-a-Service (laaS) providers. Such reserved contracts offer the consumer a significant price reduction compared to pay-per-hour pricing models, in exchange for an upfront payment. We present an algorithm that uses load prediction to make cost-efficient purchasing decisions, and evaluate whether the use of automated time series forecasting proves useful in this context. The algorithm takes into account a wide range of contract types as well as the organisation's current contract portfolio. We evaluate the effectiveness of different contract renewal policies and load predictors based on ARIMA, Holt-Winters and exponential smoothing techniques, and compare these with the performance of a simple predictor. We adopt a large set of 51 real-world web application load traces to evaluate the performance and scalability of our algorithm through simulation. Our results show that the algorithm is able to significantly reduce laaS resource costs through automated reserved contract procurement, but that the use of advanced prediction techniques only proves beneficial in specific cases. The algorithms scalability is shown to be sufficient for its adoption in settings with a large number of contracts. (C) 2015 Elsevier B.V. All rights reserved.