Generation And Optimization Of Auto-scaling Rules For Cloud Computing

Cloud computing is one of the hotspots in current study. It is emergingas a novel computing paradigm that aims at delivering computing power as autility. The vision that people will have network access to unlimited numberof computing resources is becoming more and more realistic and promisingwith the advances in related technologies. Many big players in IT industryhave invested a large amount of capital in cloud related area and manysuccessful practices come out ever since.Elasticity, also known as auto-scaling ability, is as one of the greatestadvantages cloud computing has over others. It is also the vital feature thatenables cloud platform to adaptively allocate and release resource for thedeployed systems according to their workload. The quality of service (QoS)for the target system thus is guaranteed and utilization of the resources is alsopossibly maximized. A good resource requirement model of the target systemis crucial while implementing the elasticity of cloud computing. Cloudplatform normally carries out its resource provision operation according to aspecified model or standard. A good model will enable cloud platform tomake an accurate estimation about how many resources target system needsin a certain situation, which in turn will result in great elasticity. Currentresearch work mainly focuses on real-time and on-demand optimization ofrelated parameters or virtual resource quotas for target system. But thesesolutions have some shortcomings, as they may probably bring inperturbation. This paper proposes a neural network based method frameworkto implement elasticity for cloud platform. In the proposed framework,resource requirement model is constructed through learning from the historydata of the scaling operations and their corresponding context. A neuralnetwork with reasonable structure is designed to discover the pattern behindthe history data. The eventual well-trained neural network represents a goodresource requirement model and auto-scaling rules are generated from the model thereafter to instruct cloud platform in its resource provision operation.Besides, this paper also proposes a mechanism to optimize the auto-scalingrules so that the effectiveness of the rules is guaranteed.The usefulness and effectiveness of the proposed method framework areverified through several simulated experiments. A common static scaling ruleis used as the reference line to verify the performance of the generatedauto-scaling rules. Experimental results show that the auto-scaling rulesguarantee the QoS of the target system in more than93%of time, comparedwith QoS satisfaction only in86%(or below) of situations for static rule. Theusefulness of proposed rule optimization mechanism is also verified in theexperiments.This paper considers IaaS pattern of cloud platform and takes themulti-tier web system as the representative of the target system.