Introduction

More and more businesses are looking to Linux as a possible platform for their database servers. The low cost of maintaining a Linux server makes it appealing to small- and medium-sized businesses, and even to enterprise businesses that want to reduce operational costs. Improvements in the kernel's reliability and performance, better hardware and device driver support, and an abundance of third-party tools to help businesses maintain their servers make Linux a viable and effective solution for database systems. (A distinction is made here between a database server, which is vendor-specific software that maintains a database, and a database system, which is the incorporation of the operating environmenthardware, operating system, libraries, and so on).

The enterprise database server requires a tremendous amount of system resources. Because a database must be capable of handling thousands of concurrent user connects and processing terabytes of data, it requires a high level of parallel processing power, a large amount of memory, and very large I/O bandwidth and throughput. As such, its performance is highly dependent on how effectively the kernel manages these resources. Inefficiencies in this area can lead to excessive time spent in kernel space. Also, certain inefficiencies can cause lock contentions, which can lead to poor scalability and serialization problems. Both of these behaviors can severely impact database performance because they either consume processor cycles that would normally be used by the database or force the database applications to serialize on a kernel resource.

We'll discuss various aspects of database server performance and tuning, including the following:

• An overview of database architectures

• Tunable areas of databases that have an impact on performance

• Tunable areas of the Linux kernel that have an impact on database performance