Research Of Cloud Computing Based On The Shared-nothing Architecture Parallel File System

Cloud computing is a new computing model, cloud computing provides thelarge-scale computing resource to the user in the form of reliable service by theeffective integration and multi-level virtualization to the interconnected computingresources, So cloud computing can liberate the users from the complexities of theunderlying hardware logic, software stack and network protocols, the users only needto focus on the business logic. Cloud computing has been a hot at home and abroadsince2007. Currently, Google, IBM, Amazon, Microsoft and EMC, etc has launchedtheir own cloud computing solution, meanwhile they also study in-depth the cloudcomputing in academia.The rapid development of cloud computing technology makes more and moreindividuals, enterprises accept the changes brought by it. Many enterprises applycloud computing and they have been successful. Cloud computing has brought aconsiderable cost reduction and efficiency improvement. However, with more andmore extensive and in-depth applying cloud computing, users have higherrequirements to the cloud computing about performance, stability, security. That is,users have higher requirement to the infrastructure including distributed file system,etc building cloud computing. Therefore, both the commercial and academia researchthe infrastructure, especially distributed file system.The content of this study is an innovative project on IBM. Making use of theshared-nothing architecture parallel file system (GPFS-SNC) build a simple cloud tooffer the services of cloud storage and virtual machine application. The purpose ofthis project is to explore how to make use of the features and advantages of theGPFS-SNC to build high performance, high stability, high scalability cloudcomputing. The following is the main works of this paper and the main results of thispaper. The main works are:First, this paper needs to research and analyze the architecture and features of theGPFS-SNC. This research can make us build the cloud computing which can providehigher performance, higher stability, higher scalability services of cloud storage andvirtual machine application.Second, this paper needs to research and analyze the migration and cloningtechnology which are supported by kvm, also reasearch and analyze the advantagesand disadvantages of the various types of the migration and cloning techniques;finally, this paper needs to selecte the appropritate virtual machine migration andcloning technology by combining the characteristics of the GPFS-SNC file system.Third, this paper needs to research and analyze the framework and source codeof freeNAS, because this project is secondary development based on the FreeNAS.This research makes us better port the GPFS-SNC to the freeNAS and combine theKVM to GPFS-SNC to provide better virtual machine application service.Fourth, this paper needs to research and analyze the load balancing algorithm.This paper needs to design and implement approriate load balancing algorithm bycombining the characteristics of the GPFS-SNC file system.The main results are:First, this paper achieves the technology of the interruptted web service. thissystem is to provide a web interface for users, this system still can provide webservice for users, when web server node in the cluster is down. This feature mainlymakes use of the fault-tolerant technology of the GPFS-SNC. When the web server isdown, the web service will be migrated automatically to other node in the cluster. Thetest results show that the migration time is very short, and the users do not almost feelthat the web service had been interrupted.Second, this paper achieves the technology of the load balancing to the servicerequests which are used to access their own storage space. This system provides CIFS(Samba), FTP, NFS service for users to access their own storage space. This systemachieve load banlancing to the users requests of these service to access their ownstorage space, that is, the multi-user request services are balanced distribution of eachnode in the cluster. The test results show that the request services are balancedbasically on each node in the cluster.Third, this paper designs and implements a suittable load balancing algorithm.This system uses a centralized control and polling algorithm, but the period T isintroduced to the polling algorithm. According to the information collected per period, the nodes are sorted in ascending order accord depending on the load, and the order ofnodes is stored in the load table that is in the central node, then, the first task isassigned to the first node in the load table, and the next task is assigned to the nextnode in the load table. Meanwhile, when the period T is end, the algorithm needupdate the load table, then, start to poll from the first node in the load table. This loadbalancing algorithm makes use of the unique features of GPFS-SNC; users canspecify optionally the three nodes in the cluster to store the copies of the specified file.This unique feature makes us more easily achieve a high-perforcemance loadbalancing. The test results show that this load balancing algorithm have better efficacyand stability.Fouth, this paper achives to port the GPFS-SNC to freeNAS, the snapshot androllback to the users storage space, and to create virtual machines by the way oflined-clone.