Impact of Workload and Renewable Prediction on the Value of Geographical Workload Management

There has been increasing demand for energy sustainable and lowcost operation in cloud computing. This paper proposes dynamic Geographical Load Balancing and energy buffering management (GLB) to achieve these goals which (i) shifts workload (particularly peak workload demand) toward Data centers that offer low utility rate or green energy at a time, and (ii) banks excess green and low-cost energy to shift peak workload demand away from high utility rate. Such a scheme needs to be aware of the workload intensity and the available renewable power of the cloud in future (over a relatively long prediction window such as a day). Existing solutions mainly focus on developing algorithms and demonstrating the cost efficiency of GLB, disregarding the prediction accuracy of the workload and the renewable power. However, erroneous information decreases the efficiency of GLB. This paper studies the performance of the online GLB solution when using time-series based prediction techniques (e.g., ARIMA ) for the workload and the renewable power (i.e., solar and wind). The results of the simulation study using realistic traces highlight that GLB with and without prediction error is effective in reducing average energy cost and increasing sustainability of data centers. Further, GLB is shown to be significantly effective in shaving peak power draw from the grid (e.g., reducing peak power upto 100%), however the erroneous information due to the prediction error adversely affects its performance. Furthermore, the simulation study indicates that the optimal mix of the renewable power (i.e., wind and solar) to be leveraged by GLB, is achieved when data centers are powered from both the solar and the wind power.