A Comparative Evaluation of Algorithms for Auction-based Cloud Pricing Prediction

Nowadays, cloud computing providers offer idle resources through an auction-based system in order to maximize resource utilization and economical revenue. Cloud computing consumers have the opportunity to take advantage of the resources offered at very low spot price in exchange for lower reliability in the provision of these resources. In this context, the Spot Price Prediction (SPP) is a well studied problem mainly formulated as a time series prediction, with particularities of auction-based cloud markets. This work presents a comparative evaluation of three different well-known prediction algorithms, applied for the first time to the SPP problem, against a state-of-the-art Three-Layer Perceptron (TLP) algorithm. In order to measure the accuracy of the evaluated algorithms, the following error metrics were considered: (1) Mean-Squared Error (2) Maximum Positive Error and (3) Mean Positive Error. Experimental results indicate that the Support Vector Poly Kernel Regression (SMOReg) algorithm outperforms other evaluated algorithms for the SPP problem, improving probabilities of obtaining resources in a highly dynamic spot market.