Towards Virtual Machine Energy-Aware Cost Prediction in Clouds

Abstract

Abstract. Pricing mechanisms employed by different service providers significantly influence the role of cloud computing within the IT industry. With the increasing cost of electricity, Cloud providers consider power consumption as one of the major cost factors to be maintained within their infrastructures. Consequently, modelling a new pricing mechanism that allow Cloud providers to determine the potential cost of resource usage and power consumption has attracted the attention of many researchers. Furthermore, predicting the future cost of Cloud services can help the service providers to offer the suitable services to the customers that meet their requirements. This paper introduces an Energy-Aware Cost Prediction Framework to estimate the total cost of Virtual Machines (VMs) by considering the resource usage and power consumption. The VMs' workload is firstly predicted based on an Autoregressive Integrated Moving Average (ARIMA) model. The power consumption is then predicted using regression models. The comparison between the predicted and actual results obtained in a real Cloud testbed shows that this framework is capable of predicting the workload, power consumption and total cost for different VMs with good prediction accuracy, e.g. with 0.06 absolute percentage error for the predicted total cost of the VM.

Energy-aware cost prediction and pricing of virtual machines in cloud computing environments

“…Further, an experimental study was carried out to investigate the effect of the resource usage (CPU, RAM, disk and network) on the power consumption. The findings [13,14,15] show that the CPU utilisation correlates well with the power consumption, which is supported in other work, for example [11,16,17]. Thus, the work introduced in this paper follows the same approach and takes into account the CPU utilisation only when modelling and identifying the energy consumption for the VMs.…”

“…As depicted in Figure 4, this framework includes five main steps to predict the VMs workload and power consumption, then estimate the total cost of VMs. To achieve this aim, the following steps are required [15].…”

A Categorization of Power Measurement and Power Consumption Models in the Telco Cloud

“…However, the statistics also show that some research is seeking alternatives to linear regression as a means of fitting relationships between power and its determinants. Non-linear models such as Polynomial Regression, Multi-Gaussian Regression and Exponential Regression are also being used [50,76,88,142]. More complex processors show non-linear behavior when power management is exercised.…”

“…The arguments posed by each group against the other's approach can be summarized as follows. The "system metrics" group claims that the "event counters" group's work is (a) not portable (at least across microarchitecture families) and (b) cannot be exercised without low-level access to the host (thereby, this approach cannot be exploited by user-level privileges) (see, e.g., [50], p.121 and [87], p.43). The "event counters" group claims that CPU utilization is a workloaddependent predictor (see, e.g., [72], p.1380) and therefore cannot be used without retraining the model.…”

Performance and Energy-Based Cost Prediction of Virtual Machines Auto-Scaling in Clouds

“…In this paper, we extend our work [11] by taking the performance variation into account and introduce a new Performance and Energy-based Cost Prediction Framework. This framework supports decision-making regarding auto-scaling cost while at the same time being aware of the impact on other quality characteristics such as energy consumption and performance of the application [12].…”