On-demanded Resource Prediction And Optimal Resource Allocation Method Research In Cloud Computing Environment

Cloud computing is a novel Internet-based information resource service system that can providecustomizable and flexible virtualized resources services to users，including infrastructure, platformsand applications. In the co-driver of technological progress, demand for lead and service modeinnovation and other factors, the cloud computing has been widely recognized by industry andacademia. It formed many new creative industries in real life that covering the mobile Internet,Internet of things, etc. Its low cost and ubiquitous applications make it develops rapidly, and it willchange all aspects of people’s lives radically. In order to satisfy these diverse and massive applicationresource requirements, cloud computing clusters must keep huge resources. These resources aregeographically distributed, within heterogeneous types, and have different resource managementstrategies and resource usage pricing guidelines in their respective management domains. Resourcemanagement is one of the core issues for cloud computing, its purpose is to adopt virtualizationtechnology to shield the underlying resource heterogeneity and complexity, which make massivedistributed resources form a unified giant resource pool. And on this basis, it can achieve efficientresource allocation and using by rational implementing resources management methods andtechniques. Therefore, how to achieve effective management of cloud computing resources becomesa challenging research topic. In this paper, standing on the cloud computing infrastructure operatorsand service providers’ perspective, we focused on the field of optimal resource management in cloudcomputing, including resource description, organization, discovery, matching, configuration andmonitoring, etc., and placed our research emphasis on proposing dynamic short-term loadsprediction methods, short-term loads prediction based optimal resource configuration methods, andcombined middle and long term resource prediction model construction. The aim of our research isto make cloud computing resources effective organized and reasonable configured, that ensuring thequality of service while reducing data center energy consumption and operating costs, improving theprofit of cloud computing infrastructure operators and service providers, achieving green computing,providing theoretical reference for a healthy and sustainable development of cloud computing.Based on the above discussion, in this paper, the main research contents of on-demandedresource prediction and optimal resource allocation methods in cloud computing environment are:comprehensive study of resource management in cloud computing environment, resources featureextraction and classification based dynamic short-term load prediction methods, short-term loadprediction based optimal resource allocation methods, and combined middle and long term resourceprediction method that can provide decision support for cloud computing infrastructure operatorsand service providers’ middle and long term resource capacity plan.The specific research contents and innovative work in this paper are the following aspects:Firstly，this paper summarizes the previous works in cloud computing resources descriptionformat and language, discover architecture and technology, dynamic organization, optimal allocation,real-time monitoring and other aspects of research results. We also elaborated the new problems ofresources management that need to be resolved in cloud environment. Based on these work, we builda resource management framework of cloud computing and discoursed its application inmanufacturing.Secondly, this paper analyzed the differences of load characteristics between cloud computingand grid computing, distributed computing or other high-performance computing, and then studied the important role played by on-demanded short-term resource load prediction in resources efficientallocation and optimal management. We focused on the short-term load forecasting problem in cloudcomputing and explained its important role in achieving real-time control of resources, maintainingstable operation of the whole system, reducing data center energy consumption and protecting theQoS of cloud services. Then, we built a multi-step load prediction framework that based on loadsubsequences extraction, load subsequences classification and future load forecasting. First, weadopted the fixed size overlapping sliding window to extract load subsequences from historical timeseries data. Second, we employed the kernel fuzzy c-means based supervised clustering algorithmand HMM based unsupervised clustering algorithm to identify groupings of cloud computing loaddata from a large set of historical traces to concisely represent the system’s behavior. The last step isutilizing a genetic algorithm optimized Elman network to forecast required load in next time period.By this hybrid prediction approach, an ideal result can be achieved for cloud computing loadprediction.Next, On the basis of short-term load prediction, we constructed a load prediction-basedoptimal cloud resources allocation framework, and proposed a resource monitoring and loadprediction based elasticity adaptive control system for cloud computing resources allocation. Thiscontrol system can implement a mixed resource allocation strategy that combines proactive controland reactive reaction to realize the effective use of cloud resources. Subsequently, in view of lowresource utilization problem caused by the single virtual machine for single customer (SVSC)resources allocating mode that most present cloud service providers adopted. In this paper, we firstbuilt a novel public cloud infrastructure with five layers and then proposed a single virtual machinefor multiple customer (SVMC) resource allocating strategy to enhance the resource utilization inpublic cloud environment. According to the various resource requests from different cloud serviceusers, SVMC resource allocating strategy is able to search for the optimal virtualized resourcesautomatically, and can allocate different application requests into the same virtual machine withoutaffecting the quality of service. As a result, this resource allocating strategy makes cloud computingproviders to ensure quality of service while improving efficiency in the use of cloud computingresources, and reduce energy consumption.Finally, in terms of demand for middle and long term resource management in cloud, andaccording to the load characteristics of middle and long load characteristics in cloud computing, weconstructed a generalized fuzzy soft sets theory based middle-long term load combinationforecasting model for cloud computing, and proposed a novel angle cosine based similaritymeasurement of generalized fuzzy soft sets. Then we adopted the similarity measurement result andthe prediction accuracy from ANFIS (Adaptive Neuro-Fuzzy Inference System) and seasonalARIMA model to obtain the weights of combined prediction model. On this basis, we constructedthe generalized fuzzy soft sets theory based combined forecasting model GFSS-ANFIS/SARIMA.