Proactive Workload Forecasting Model with Dynamic Resource Allocation for Modern Internet Application

Abstract

Modern Internet applications are subject to significant variations in workload demand and this may affect the applications performance. Such kind of applications are usually hosted on multi-tiered, cluster based web hosting environments. Dynamic resource allocation play a curial role in handling sudden events where servers are moved from another (quieter) pool to meet such demand. In this work two well-known dynamic switching policies -- the Proportional Switching Policy (PSP) and the Bottleneck Aware Switching Policy (BSP) -- alongside the proactive properties of a workload forecasting model -- Simple Moving Average (SMA) -- with several different interval times. The experiments have been conducted over a real-time Internet traces. The results show that changing the interval time alongside the workload forecasting model can be very effective when applied alongside dynamic resource allocation strategies. The improvement of the system performance can be up to 12.4% when the right combination are conducted between the proposed approaches.