A Crisis Mitigation Policy for SLA Profit Maximization in Multitenant Database

SLAs (Service Level Agreement) are agreements negotiated by the multitenant database service providers and their tenants in order to measure the quality of database service. They rule the fee paid to the service providers if the SLAs are satisfied, and the penalty by the service providers if their database service violated the SLA constraint. In fact, resources are shared among different tenants in a cloud multitenant environment, and tenants usually have changeable workload pressure over time. To maximize the profit, it is important for database service providers to intelligently managing the usage of resources. This paper focuses on managing tenants´ workloads for fixed resources of a distributed database. The system uses historical operation data of every tenant to learn a model that describes the resource consumption for each tenant. Based on the learned model the crises detecting module periodically collects the free resources state of every node and judge if it has crises. If crises are detected, the crises mitigating model plays its role by leveraging a revised sliding window algorithm to find a rational target node, the following workload of the crisis tenant will be dispatched to the target. Extensive experimental results demonstrate the effectiveness and efficiency of our crises mitigation policy.