Research On Elastic Dynamic Balance Issue Of Facing The Complex Load Characteristics And Performance Requirements Of Cloud Database

With the development of cloud computing, more and more traditional applications began to migrate to the cloud platform. One of the main characteristics of cloud computing is the ability to expand elastically, the so-called flexibility is the ability to dynamically increase or reduce resources when the load of system change dynamically. Meanwhile cloud database elastic dynamic balance issue has always been thought is the key to realize the cloud computing elastic characteristics. From the tenant’s perspective, they want their performance to get 100% satisfying and the price of cloud services is inexpensive. From the cloud service provider’s perspective, they hope the performance of the tenants can meet and the resources used the least. If we are able to accurately predict performance of tenants and resource usage of data node, we can provide a good guidance for the elastic dynamic balance strategy accurately of cloud database. However, due to the complexity of the tenant’s performance and workload under cloud environment make the cloud database elastic dynamic balance problem becomes more complex, and amplify the technical difficulties of implementing adaptive flexible control of cloud database system.In cloud computing environment, the applications of Multi-tenants often exhibit complex mix of high dynamic load characteristics. High dynamic is mainly reflected in the large fluctuation, due to different tenants superposition of different load fluctuations are amplified. The Mix embodied in different tenants have different load special permit, but the overall load is not a simple superposition. And the heat and peak data of different tenants are different, so that the performance model of tenants and the node resource model building becomes complicated.Cloud computing environment is usually shared tenant data storage, thus multi-tenant resource competition will exist. Competition for resources increases the complexity of the tenant performance and the difficult of resource usage prediction. Multi-tenant cloud resources are shared storage, so it may happen that load high load low tenant will occupy the resources of the tenants or a tenant because the load characteristics of the changes and make changes in the use of resources. Then, the fair of the tenants were broken because of the competition of the resources. It is not a simple linear relationship between the performance of tenants, relational data node resource usage and load.In cloud computing environment, the performance of Multi-tenant is not the same, and the complexity which increasing the difficulty of cloud database node data distributed data placement, is brought by multi-tenant personalized performance requirements. The complexity of cloud application performance of needs is mainly reflected in the performance of the different needs of different tenants. Due to the same node often needs to store data of multiple tenants, split or merge a group of tenant data placed into the appropriate node to determine whether it can meet the performance requirements of this group of tenants for application data becomes extremely difficult.In this paper, a systematic study of the elastic dynamic balance problem for key/value NoSQL type of cloud data storage system whose mode is shared, and propose an elastic dynamic balance control framework, which includes tenant performance model based on Gaussian Process. The model was able to predict the performance of multi-tenant whose properties with complex features accurately; including the node resource model based the approach of KCCA, using the model can accurately predict the resource usage in the case that existing fierce competition for resources data nodes for resource sharing models; finally, the generating algorithm of elastic dynamic balance strategies based on the model of tenants’ performance and the model of node resource is introduced. The node resource utilization is high in the case of tenant properties met this node by executing the algorithm, while the whole system overhead is minimized.The model and algorithms proposed in this paper are validated the accuracy and validity through experiments in multiple aspect.