Design And Implementation Of Multi-Tenancy File Storage System For SaaS Applications

Software-as-a-Service, cloud computing, internet applications, scientific computing and other modern large-scale application platforms demand some requirements while using the file storage system, such as greater capacity, higher performance, higher reliability and multi-tenant isolation. Efficient and credible mass file storage will be the key technology to build such large-scale applications.For multi-tenant SaaS applications, data isolation and quality of service guarantees, this paper presents multi-tenant-oriented mass storage system architecture. Metadata server and storage server clusters are used to separate file system namespace operations and data stream operations. With virtual storage controllers in metadata server, multi-tenant's requests are authenticated for security. Metadata cache is used to improve the performance of concurrent access to metadata. Caching strategy reduces the delay of server failure recovery time. Storage server uses object approach to maintain data blocks. Section is the unit to allocate disk space. It makes the distribution of data blocks on disk more continuous and enhances the efficiency of large file access.In order to ensure high availability of SaaS application, this paper proposed a dynamic replica management strategy based on load balancing. Replicas are partitioned into different clusters. Super nodes, which are steady online and have high processing capacity, are set up to maintain replica addition and replica removal in clusters. Replica locate speed is accelerated by this replica organization model of centralization in local and peer to peer in wide area. Large-scale replicas can be managed effectively in this way. By analyzing the history of file access records, file access frequency in next cycle can be predicted. Replicas are added for prevalence files, and then client requests can be dispatched according to server load information to achieve a distributed system load balancing. Load migration algorithm is used to achieve both load balance and storage space balance on each node, which can form a reasonable distribution of data.In order to achieve performance isolation and service-level distinction for multi-tenant SaaS application, this paper proposed a service-based classification of multi-class service level queue scheduling method. Queue is used to manage different user's requests. Large request are break up in order to make full use of storage and network parallelism resources, and to facilitate QoS guarantees. Tenant's requests are classified according to service level. It allows the system to meet the demand of high level requests. If there are sufficient server resources, every level's QoS requirements will be satisfied.As a major national research project, we implement a mass storage system prototype JTangFS for multi-tenant SaaS applications. JTangFS implements the basic functions described in papers. It has been tested and verified by SaaS-based network disk system and Zhejiang Telecom BizNavigator platform. System test report shows that JTangFS is able to meet current demands of SaaS applications.