CloudScale: Elastic Resource Scaling for Multi-Tenant Cloud Systems (Essay Sample)

CloudScale: Elastic Resource Scaling for Multi-Tenant Cloud Systems
Presented by
Author Names
Institution/University
Introduction
A wide range of infrastructure such as Service (IaaS) suppliers utilizes virtualization techniques in integrating applications and facilitating isolation among obstinate users (Barham and Xen, 2003; Chandra et al., 2004). Nevertheless, statically segmentation of physical resources into VMs (virtual machines) with respect to peak requirements of the application will result to meager resource exploitation. The application is subjected to SLO infringements incase the resource cap is extremely minimal. On the other hand, over elevated resource cap means that the server will incur extra charges (wasted resources). To evade both conditions, there is the need for multi-tenant cloud structures to engage elastic resource leveling approaches in order to dynamically regulate the resource cap upon the demands of the user (Chase et al., 2001). Resilient resource leveling enables cloud systems achieve SLOs (service level objectives) with least amounts of resource provisioning (Clark et al., 2005). It was on such grounds that CloudScale was instigated. The system utilizes online resource demand envisaging and prediction fault managing to attain adaptive resource distribution devoid of prior knowledge assumption concerning the applications operating within the cloud (Hsu and Fang, 2005).
Prediction of temporary resource requirements is achieved through feeding the resource process duration series into an online resource command envisage model. This prediction model utilizes a fusion technique that uses state-driven and signature-driven prediction criterion in realizing both low overhead and high precision. Primarily, the approach utilizes FFT (fast Fourier transform) to make out the replicating outlines known as signatures. Upon identifying a signature, the prediction model utilizes it in the approximation of future resource requirements. Otherwise, a distinct-time Markov sequence is utilized by the prediction model to determine the future resource demand. Under-estimation faults are avoided by the error correction module via executing online adaptive stuffing. The reactive fault rectification constituent identifies and regulates under-estimation faults that cannot be averted in the padding scheme. The outcome is a preliminary resource cap for ...