IBM Tivoli Monitoring, Version 6.3

Understanding the disk requirements for your database

Consider the factors in this section when designing a disk subsystem to support your database processing.

Disk crash: With sufficient funds and planning, you can build a system that can continue running without interruption or be recovered in a few hours, despite a system crash. You can provide disk protection by using some level of RAID on parts, or all, of the disk subsystem. Some of the more popular types of RAID include the following:
- RAID 1, also known as disk mirroring, uses data mirroring to achieve a high level of redundancy. In a RAID 1 configuration, two copies of the data are kept on separate disks, each mirroring the other.
- RAID 5 uses block-level striping with distributed parity. RAID 5 stripes both data and parity information across three or more drives.
The database log needs disk protection to enable database recovery of recent transactions. The other disks can optionally be protected. If they are protected, you can eliminate downtime while data is recovered.
Consider the OS and paging space disks.
Also, consider including one or two additional disks to speed up recovery and reduce the risks while running after a disk failure.
Because of the increasing capacity of disk drives, the configurations listed below result in excess disk capacity but increase the number of disks available for I/O throughput.

The following table provides some example sizes for a database:

Table 1. Database size examples
	Number of disks to use
	Absolute minimum disks	Small RDBMS	Small and safe RDBMS	Large RDBMS
Operating System	1	1	1 + mirror	1
Paging and RDBMS code	Use above	1	1 + mirror	1
RDBMS data	1	1	1 + mirror	8
RDBMS indexes	1	1	1 + mirror	6
RDBMS temp	Use above	1	1 + mirror	6
RDBMS logs	1 + mirror	1 + mirror	1 + mirror	2
Database data	12 GB	24 GB	48 GB	108+ GB
Number of disks	5	7	12	24

The Absolute minimum disks column specifies the minimum number of disks for an RDBMS. In this column, the index and temporary space is allocated onto one disk. While not an ideal arrangement, this might work in practice because databases tend to use indexes for transactions or temporary space for index creation and sorting full table scan large queries, but not both at the same time. This is not a recommended minimum disk subsystem for a database, but it does have the lowest cost.

The Small RDBMS column represents a minimum disk subsystem, although there might be limits in I/O rates because of the data being placed on only one disk. Striping the data, indexes, and temporary space across these three disks might help reduce these I/O rate limits. This disk subsystem arrangement does not include disk protection for the database or other disks (apart from the mandatory log disk protection for transaction recovery).

The Small and safe RDBMS column adds full disk protection and can withstand any disk crash with zero database downtime.

The Large RDBMS column represents a typical size database for a database subsystem. Disk protection is not included in these sizings but can be added to increase the stability of the database.

Feedback