Dynamic Resource Allocation and Active Predictive Models for Enterprise Applications

This work is concerned with dynamic resource allocation for multi-tiered, cluster-based web hosting environments. Dynamic resource allocation is reactive, that is, when overloading occurs in one resource pool, servers are moved from another (quieter) pool to meet this demand. Switching servers comes with some overhead, so it is important to weigh up the costs of the switch against possible system gains. In this paper we combine the reactive behaviour of two well known switching policies – the Proportional Switching Policy (PSP) and the Bottleneck Aware Switching Policy (BSP) – with the proactive properties of several workload forecasting models. Seven forecasting models are used, including Last Observation, Simple Algorithm, Sample Moving Average, Exponential Moving Algorithm, Low Pass Filter and Autoregressive Moving Average. As each of the forecasting schemes has its own bias, we also develop three meta-forecasting algorithms (the Active Window Model, the Voting Model and the Selective Model) to ensure consistent and improved results. We show that request servicing capability can be improved by as much as 40% when the right combination of dynamic server switching and workload forecasting are used. As important is that we can generate consistently improved results, even when we apply this scheme to real-world, highly-variable workload traces from several