Turbocharging DBMS buffer pool using SSDs
Proceedings of the 2011 ACM SIGMOD International Conference on Management of data
Data management over flash memory
Proceedings of the 2011 ACM SIGMOD International Conference on Management of data
| Hi-index | 0.00 |
Technology is always changing. In three years, today's fast computer will be outdated, and, if not already obsolete, it will be fast on its way. This dissertation focuses on two hardware trends, processors getting faster and more powerful and the introduction of solid-state disks, and explores ways to take advantage of them in order to improve database performance. Processors have improved at a much faster rate than have memory and disk bandwidths. One way to overcome this difference is to use compression, which trades processing cycles for memory bandwidth to improve system performance. The first chapter explores the trade-offs in how much compression a system should use versus how much bandwidth the compression saves. The second chapter explores two different paradigms for storing data physically on disk, row stores and column stores, and how their performance compares for different queries and tables, when the data for both paradigms are compressed. The next chapter focuses on how to leverage a solid-state disk's different read and write characteristics (compared to hard disk). We introduce the Solid-State Disk Buffer Manager (SSD-BM), a novel scheme where a solid-state disk is used as a secondary buffer pool. The pages can be accessed quickly thanks to the solid-state disk's fast random reads, and they are only written sequentially, to overcome the solid-state disk's sometimes poor random write performance.