Research And Implementation Of A Resource Management Framework For Private Clouds Based On SLA

Resources management in cloud computing environment includescomputing resources, storage resources, and network resourcemanagement. Compared to public cloud, computing resource in privatecloud computing are limited, computing resource espacially virtualmachine resource management issues are important. This work is based onthe background of an actual project of cloud computing platform of aprovince, according to the actual needs of the project, in its private cloud,there’re such problems of virtual machine resource management:1) it is acontradiction between the scalability of the application and the limitedvirtual machine resources. How to ensure the level of service according toservice level agreements (SLA) and allocate the virtual machine resourcesto applications reasonably.2) In order to take full advantage of theresources, the sum of virtual machine resources allocated to multipleapplications by SLA in one server may larger than the physical resourcesof the server. An application’s load increase may lead to seize the resourcesof other applications. How to solve such a performance isolation problem.To solve these problems, this work designed an algorithm of virtualmachine resource request admission control based on SLA and analgorithm of virtual machine performance isolation based on predictionand dynamic programming. On this basis, the work describes the designand implementation of a resource management framework for privateclouds (RMFPC) based on SLA. The result of the test shows theframework is feasible and effective.The main contents of this thesis include: 1) According to the first problem, this work designed an algorithm ofvirtual machine resource request admission control based on SLA. Thealgorithm consists of two parts; one is to directly allocate virtual machineresources to the resource requests which meet the conditions and put thoseother requests to the requests waiting queue; the other one is to responseapplication exit events, allocate virtual machine resources to some requestsin the requests waiting queue reasonably.2) According to the second problem, this work designed an algorithmof virtual machine performance isolation based on the prediction anddynamic programming. The algorithm predicts the resources seizing ofvms by curve fitting and determines whether need to migrate vms; thendetermines the target host to migrate vms by dynamic programming.3) On the basis of the above work, designed and implemented aresource management framework for private clouds (RMFPC) based onSLA. In addition to implement the core modules of the two algorithms, thisframework also includes host and vms monitoring interface, virtualmachine operating interface, the realization of these interfaces can beadapted to the current main commercial products of virtual machinemanagement.4) The test of the framework includes functional testing andperformance testing. The functional test of virtual machine resourcerequest admission control was finished by four analog applications. Theresults show that the framework achieved the core functions mentionedabove. The test of virtual machine performance isolation was finished byincreasing the load of an application. This thesis simulated three scenesand compared the running time of a same application, the test results showthat using the algorithm is faster than without using any algorithms by50%and is faster than without using predicting algorithm by28%.