Machine learning based Model for Cloud Load Prediction and Resource Allocation

Abstract

Elasticity and the lack of upfront capital investment offered by cloud computing is appealing to many businesses. There is a lot of discussion on the benefits and costs of the cloud model and on how to move legacy applications onto the cloud platform. Here we study a different problem: how can a cloud service provider best multiplex its virtual resources onto the physical hardware? This is important because much of the touted gains in the cloud model come from such multiplexing. Studies have found that servers in many existing data centers are often severely under-utilized due to over-provisioning for the peak demand. The cloud model is expected to make such practice unnecessary by offering automatic scale up and down in response to load variation. Besides reducing the hardware cost, it also saves on electricity which contributes to a significant portion of the operational expenses in large data centers.

Proper resource allocation for various virtualized resources must be based on these cloud load predictions. The presence of heterogeneous applications, such as content delivery networks, web applications, and MapReduce tasks, complicates this process. Cloud workloads with conflicting resource allocation needs for communication and information processing further exacerbate the difficulty.