Index selection plays a substantial role in database performance by reducing the I/O cost. Existing index advisors apply different heuristic methods to search the large search space of possible attributes for indexing. These heuristic approaches do not have a mechanism to learn about the goodness of the recommended index set. Thus, they might choose the same index set with a low impact on I/O cost reduction. Learning from their decisions can improve the quality of the recommended index set. We believe that Deep Reinforcement Learning (DRL) is a solution to tackle this issue. Using DRL, an index advisor can improve its decision using the feedbacks of its decisions. In this paper, we propose a DRL-index advisor for a cluster database. We describe the major components such as agent, environment, set of actions, the reward function, and other modules. We conclude the paper with open challenges and possible future work.
|Original language||English (US)|
|Title of host publication||Proceedings - 2020 IEEE 36th International Conference on Data Engineering Workshops, ICDEW 2020|
|Publisher||Institute of Electrical and Electronics Engineers Inc.|
|Number of pages||4|
|State||Published - Apr 2020|
|Event||36th IEEE International Conference on Data Engineering Workshops, ICDEW 2020 - Dallas, United States|
Duration: Apr 20 2020 → Apr 24 2020
|Name||Proceedings - 2020 IEEE 36th International Conference on Data Engineering Workshops, ICDEW 2020|
|Conference||36th IEEE International Conference on Data Engineering Workshops, ICDEW 2020|
|Period||4/20/20 → 4/24/20|
Bibliographical notePublisher Copyright:
© 2020 IEEE.
Copyright 2020 Elsevier B.V., All rights reserved.
- Cluster computer
- Deep reinforcement learning
- Index selection