Research On Workload Modeling And Resource Prediction In Cloud Computing

Recently,there is a huge waste of cloud spend when cloud users rent public clouds.So utilizing cloud resources on demand has become the focus of the industry concern.And the simulated workloads are generally used to predict the demand of cloud resources.The existed workload models were always for specific application or a single type of application,as well as covered the insufficient diversity of workload variations.Besides,the accuracy of resource prediction based on workloads was restricted to a specific workload and the type of cloud resources,which cannot adapt to the diverse workloads of cloud users.For that,the dissertation proposed an abstract model of cloud application workload and a corresponding instantiation approach to mimic “the real workload”.On this basis presented the resource prediction methodology to answer the question “how to determine the needed resources according to the workload”.The main innovation points of the dissertation are as follows:(1)An abstract hierarchical model of cloud application workload is proposed.It consists of an abstract hierarchical model of cloud application and an abstract hierarchical model of workload variation.The abstract model of cloud application is built with a unified mathematical method to describe different types of cloud applications.Meanwhile the abstract model of workload variation is built by extracting the substantive characteristics of workload variations,which contains an abstract model of arrival process and a three-hierarchies workload variation model from user,application to service units.This work establishes the theoretical basis of generating the workloads with diversified variations and hybrid applications.(2)An instantiation approach of the abstract model of cloud application workload and an adaptive workload generation algorithm are presented.The instantiation approach instantiates the abstract model of cloud application workload into different types of workload models.The adaptive workload generation algorithm then adaptively generates workloads according to the different types of instantiated workload models.Both of them together provide a technical route from the abstract model of cloud application workload to hybrid diverse workloads.(3)A cloud resource prediction approach based on the abstract model of cloud application workload is presented.The approach takes generated workloads based on the instantiated workload model as input,builds a multi-dimensional cloud resource model,enriches the critical influences on resource prediction during the Profile phase,and presents a cloud resource prediction algorithm which is independent of application types,workload variation patterns and cloud resource types.The experimental results show that the prediction accuracy reaches 90.5% on average.On the basis of above theoretical contributions,the dissertation designs and develops the system for adaptive workload generation and performance monitoring,Cloud-WG.It supplies interface for cloud users to define their own application and the possible workload variations to be tested,then generates automatically the corresponding workloads,monitors the performances of cloud resources at runtime.Cloud-WG provides a reference for cloud users in their use of cloud resource.In sum,the dissertation presents a complete solution for “utilizing cloud resources on demand” in theoretically and practically.