| With the development of cloud computing and the mature of the applications, SaaS (Software as a Service), as an important form of cloud computing, gets more and more attention. It has gradually become an important way to apply advanced technology for Small and Medium Business (SMB). Delivery platform for SaaS application promotes the vigorous development of the SaaS mode, more and more individuals or institutions rent various kinds of applications through SaaS platform. These applications cover all areas of people's lives, through applications and collaborative work between them to complete some things, including retrieval, transaction management and analysis, etc.Currently, most matured SaaS service provider use single-instance multi-tenancy software delivery model, where thousands of tenants share one application and business data is stored in a shared database, thus the end users feel the instance is exclusive to him and not aware of the existence of other customers at the same time. The customizable storage architecture realizes the resource share at hardware-level for multi-tenant's. However, the number of tenants which a single data node can support is restricted by hardware, so it is hard to achieve the dynamic expansion of the scale when the tenant require more storage space or higher quality of service, only by upgrading the hardware. Along with the number of tenants in SaaS delivery platform as well as the amount of information grow exponentially curve, the data management in SaaS delivery platform is on the inevitable stage of changing from single data node to multi-data node in the cloud. Correspondingly, people expect more about the demand of data management in SaaS application delivery platform.This paper focuses on the key technology of the multi-tenant cloud data management in the SaaS application delivery platform, aiming at supporting the rapid development and delivery maximumly meanwhile guaranteeing the high-speed movement of the multi-tenant SaaS application. Multi-tenant data management in the SaaS application delivery platform has its own characteristics, which can not to be solved effectively in the current cloud data management:(1) SaaS application delivery platform involves a large number of SaaS applications, the data models vary widely in different applications of business areas; Each SaaS application has many tenants who have the semblable model while having different business areas. Even though the same tenants in different development stage, the data model also has its own characteristics. Therefore, the platform must support data model to be extended effectively, at the same time, solve the problem of performance degradation due to the increasing complexity of the data management. (2)Currently, the data management under the background of the cloud is directly to tenants, who rent the data services, but the SaaS application delivery platform, where tenant rent its application. Data models for the same SaaS application are semblable between tenants, while different applications have various relations. Therefore, it is necessary to introduce the characteristics of multi-tenant data model in the cloud data management that guarantee effectively split and migration of the cloud data. (3)Current Cloud data management can't provide multi-level, fine-grained indexing mechanism for the tenants in the SaaS application, because it establish a virtual database for every user and not aware of the application tenant in the SaaS application delivery platform; At the same time, the traditional indexing mechanism has become invalid in the local data node and then can't provide an effective logical index for the tenants, making the operations on the random data of the tenants become difficultyThis dissertation aims at the multi-tenant cloud data management in the SaaS application delivery platform and research on the key issues. The main work and contributions are summarized as follows:1. Proposing a multi-tenant virtualization approach for SaaS platform and a hierarchical data model to solve large-scale SaaS application delivery bottlenecks of cloud-based data management technology, which shielding the developers' perception to the cloud data management technique and supporting the tenant to customize its system on demand.For the demand of "shared database, single-instance multi-tenancy" in the platform and technical bottlenecks of SaaS application development, the paper establishes application layer multi-tenant virtualization approach and hierarchical data model,which can support development through standard SQL and effectively support the tenants'customization, the unified management of data, and facilitate data sharing between applications, data management rights model and transaction management, but also easy to insure the data node stretching in the cloud. The prototype system shows that the architecture has high independence, which supports development through standard SQL; Enables the perception of various multi-tenant sharing storage methods through mapping technique according to the metadata, supporting the elastic of the data node, and providing the friendly programming model, high consistency, high scalability and high availability for SaaS application delivery platform.2. A logical model of multi-tenant cloud data on the combination of the multi-sparse and key-value, as well as a metadata storage model on this basis, which effective solving the problem of redundant storage caused by multi-tenant shared sparse tables and metadata, at the same time getting higher performance.For the problems in multi-tenant storage model, including sparse data, customization limitation, metadata redundancy and etc, the paper divides data to many sparse tables to increase the intensity of sparse table, which solving the waste of storage space because of null values in sparse tables, as well as the decrease of the access performance and inefficiency of relation join operations in the SaaS platform; By the expansion storage mechanism of key-value, solve the limitations in the customization of the sparse table; By multi-level metadata storage mechanism, solve the problem of redundant storage of metadata. Experimental results show that if the number of columns in the tenant view is normally distributed, the data intensive degree could be average improved 20%, relation join efficiency will be improved along with the increase of tuples, and redundant storage of the customized data will decrease down to 56.7%. In all, it is an effective storage model.3. A multi-tenant partition model and dynamic data migration strategy in the SaaS application delivery platform, reduces the numbers of distributed transaction, simultaneously ensure the scalability of the data nodes in cloud and the overall performance of the platform.If found a virtual database for SaaS in the cloud, the data of tenant is shared stored, so the cloud database can't manage the data in every SaaS application tenant, such as partition?shifting and backup. Metadata-driven data partitioning mechanism for tenants in the SaaS platform ensure that the transactional operation for tenant can complete within a single node, and largely avoid the distributed transaction processing; Dynamic data migration strategy moving the data partition belong to the tenant in the cloud to ensure the load balance of all the data nodes and efficient operation of the platform. This paper provides a better user experience. Experimental results show that the data migration happens when data volume in the data nodes reaches the threshold value 50%, requests from tenants were not significantly affected during the migration. Access costs for tenants reduce after migration, and the overall performance of the platform has been improved.4. A multi-level, fine-grained indexing mechanism based on SaaS platform. It improve the data service performance of the platform through quickly locate to the data node belong to tenant and effectively solve some issues in local data node, such as tenant index failure and limited customization under share tables storage mode.The multi-level, fine-grained indexing mechanism based on SaaS platform including three parts:tenant node index, tenant logical index, relational database physical index. The tenant node index solves the problem that cannot random access relevant data nodes of tenant. The tenant logical index satisfies the index customization, isolation and other demands. The relational database physical index provides efficient access mechanism for logical index and guarantee higher query performance of metadata. For tenant logical index, this paper proposes the metadata-driven mapping table index mechanism based on the key-value storage mode. This model forms the index metadata for the tenant's business data, according to requirements of tenant and realizes the isolation and customization of the index data through metadata driving. This paper gives the index maintenance strategy, slicing the index according to the data requests by tenants, using the gradually thinning index slice as the base unit of data access and to return the result set quickly. Experimental results show that the index maintenance and data access with better overall performance under the condition that distribution of the data access is balanced. |
PDF Full Text Request