SLA-Aware Cloud Query Processing with Reinforcement Learning-based Multi-Objective Re-Optimization

International audienceQuery processing on cloud database systems is a challenging problem due to the dynamic cloud environment. In cloud database systems, besides query execution time, users also consider the monetary cost to be paid to the cloud provider for executing queries. Moreover, a Service Level Agreement (SLA) is signed between users and cloud providers before any service is provided. Thus, from the profit-oriented perspective for the cloud providers, query re-optimization is multi-objective optimization that minimizes not only query execution time and monetary cost but also SLA violations. In this paper, we introduce ReOptRL and SLAReOptRL, two novel query re-optimization algorithms based on deep reinforcement learning. Experiments show that both algorithms improve query execution time and query execution monetary cost by 50% over existing algorithms, and SLAReOptRL has the lowest SLA violation rate among all the algorithms