Cloud work load prediction through different models based on time-series

Abstract

Scheduling of computational load and actual processing is an important problem to be considered from the perspectives of time and consumed energy for execution in the scale of data centers. In this paper, time-series analysis of the arrivals of the workloads have been done by applying auto regression (AR), moving average (MA), auto regression and moving average (ARMA), and Holt-Winters approaches. Performances of the four methods was evaluated and compared for Google workload logs that is publicly available in the Internet.