In many settings, a database server has to be restarted either in response to a failure event, or in response to an operational decision such as moving a database service from one machine to another. However, such restarts pose a potential performance problem as the new database server starts off with a cold buffer pool. As a result, the database application experiences a dramatic reduction in performance right after the restart, since just before the restart the database buffer pool was filled with hot pages and after the restart the database buffer pool is empty. To address these issues, traditional database systems use mechanisms such as SQL Server's aggressive page expansion and MySQL's buffer pool preloading. However, these approaches have key limitations including long warm-up times, possible early hot page eviction, user query performance saturation, and failure restart. In this paper, we present a new framework for SQL Server that allows continual capturing of the state of the buffer pool, and restoring the server state quickly with a snapshot of the buffer pool at restart. Our empirical evaluation demonstrates that our method reduces the time to regain peak performance by a factor of 2X or more over the previous approaches.
|Title of host publication||2016 IEEE 32nd International Conference on Data Engineering Workshops, ICDEW 2016|
|Publisher||Institute of Electrical and Electronics Engineers Inc.|
|Number of pages||8|
|Publication status||Published - 2016 Jun 20|
|Event||32nd IEEE International Conference on Data Engineering Workshops, ICDEW 2016 - Helsinki, Finland|
Duration: 2016 May 16 → 2016 May 20
|Name||2016 IEEE 32nd International Conference on Data Engineering Workshops, ICDEW 2016|
|Conference||32nd IEEE International Conference on Data Engineering Workshops, ICDEW 2016|
|Period||16/5/16 → 16/5/20|
Bibliographical noteFunding Information:
This work was supported in part by a gift donation from Samsung and Microsoft, and by the National Science Foundation under grants III-0963993 and IIS-1250886.
© 2016 IEEE.
All Science Journal Classification (ASJC) codes
- Artificial Intelligence
- Information Systems
- Computational Theory and Mathematics
- Computer Networks and Communications
- Information Systems and Management
- Computer Graphics and Computer-Aided Design