The disk subsystem is the most critical aspect with regard to overall database performance. A well-architected disk storage design will ensure that there is enough I/O bandwidth to support peak database activity, allowing I/Os to complete within the expected amount of time. The physical channel layout and number/type of I/O controllers are important considerations. Using RAID Level 1 or 10 arrays is recommended for the most optimal storage configuration. A higher quantity of smaller, faster physical devices within an array will boost read/write access times over a configuration that uses fewer disks with higher storage capacity.
Database disk access is random or sequential. Random I/O occurs when data is accessed from different parts of the disk. Sequential I/O occurs when the disk accesses adjacent data on a disk drive. Higher throughput can be achieved when performing sequential I/O.
It is extremely important to separate randomly accessed data from sequentially accessed data.
Within an OnBase database implementation, there are three categories of file I/O to consider: transaction log / redo log, tempdb / temporary tablespace, and data files.
The transaction log and redo log are written to sequentially, and the speed of writes to these files can be a limiting factor of database performance. It is recommended that the transaction log or redo log(s) be placed on dedicated disks, or disks that can support writes in less than 10 ms, to eliminate I/O contention with other files.
Any query that generates a hit list for a user in OnBase requires space in tempdb or the temporary tablespace, and the data is accessed randomly. The tempdb and the temporary tablespace are shared resources. The larger the implementation, the more this resource is utilized.
The files that store OnBase database tables and indexes (filegroups or tablespaces) are also randomly accessed and have percentages of reads and writes that vary between OnBase solutions.
The following are purchasing considerations for database storage:
-
Use RAID 10 (Mirroring plus Striping) for arrays that will store database files.
-
Fiber-channel, SCSI, SATA, and flash-based disk drives are all acceptable, and faster RPM disks (15K+) or flash-based drives are encouraged. It is recommended that you purchase smaller, faster RPM disks over slower, higher-capacity disks.
-
To prevent disk contention between the database and other applications, database LUNs should reside on a dedicated disk. Smaller capacity drives dedicated to the database LUNs will offer higher performance than larger drives shared between applications. This is especially true for the transaction log, which generally does not utilize much disk space but is crucial to performance due to the frequency and type of access.
-
Use Caching Controllers with a battery backup. For enterprise systems, two or more controllers are recommended. It is recommended that you enable Write Back cache and allocate 90% of the cache to Write activity for all database-dedicated arrays.
Hyland Software, Inc. recognizes that implementations that rely on a SAN for database storage may not have dedicated disks for database files. It is possible to achieve desired performance with a SAN, depending upon the configuration (for example, utilizing a wide-thin approach or virtualized or dynamic storage) and SAN cache. Ultimately, it is necessary to have enough disk resources dedicated to both meet disk space needs and to support the I/O requirements of the OnBase database solution. For more information on values for disk latency and database I/O concepts, please see the following articles:
-
http://technet.microsoft.com/en-us/library/cc966412.aspx