A Cost-Effective Query Optimizer for Multi-tenant Parallel DBMSs.

For a multi-tenant DBMS, at any point in time, a tenant may submit an SQL query associated with a performance SLO (Service-Level Objective). Any violation in the SLO requires the provider to pay penalties. As such, the DBMS which is in charge of the provider’s long-term benefit, should find a fair compromise between the financial costs brought on by resource consumption and the SLO satisfactions that affect the provider’s income. Such compromises should be carefully considered due to the instability of multi-tenant query workload. In this paper, we propose an execution plan selection strategy which changes the optimization objective in response to the query workload. This latter information is provided by the queue manager, a new component of our extended optimizer, which controls the selected execution plans’ scheduling as well as resource allocation. Experimental results showed that it proved successful in terms of the provider’s long-term profit despite variations in the multi-tenant workload.