6.1.1 Points to consider when preparing disks

To improve HADB's processing performance, it is important to prepare disks so that I/O operations are distributed. For example, if all data is stored on one disk, I/O operations will be concentrated on that disk. If the number of I/O operations exceeds the disk's processing capability at any time, processing performance will be degraded because of the I/O operations that must wait. Therefore, distribute the data across multiple disks in order to reduce the number of I/O operations for each disk.

Especially when a large volume of data must be handled, storing the files listed in the table below across multiple disks can be an effective way to distribute I/O operations.

Table 6‒1: Files that might result in improved performance by storing them on multiple disks

No.	File or directory type	Detailed listing of files
1	DB area files	Master directory DB area file Dictionary DB area file System-table DB area file Work table DB area file Data DB area file
2	Directories and files used by the HADB server	Server directory and the files under the server directory DB directory and the files under the DB directory
3	Temporary work files	Temporary work files created when the following commands are executed: adbimport command adbidxrebuild command adbmergechunk command adbunarchivechunk command adbreorgsystemdata command
4	Input data files	Input data files used during execution of the adbimport command Input data files (CSV files) used during execution of the ADB_CSVREAD function
5	Output data files	Output data files used during execution of the adbexport command
6	Archive directory	archive file
7	Synonym dictionary file directory	synonym dictionary file
8	Unload file	Unload file created when the adbreorgsystemdata command is executed

When determining the storage locations for the files shown in the table, it is best to use disks in a RAID array. If there are multiple RAID groups (groups of disks configured in RAID arrays), you can distribute disk I/O operations more efficiently by employing one of the following techniques:

• Use of storage functionality to perform distributed placement of data on virtualized disks

• Striping configuration that uses a logical volume manager (LVM) supported by the OS

■ Example 1

The following figure shows an example of using storage functionality to perform distributed placement of data on virtualized disks.

Figure 6‒1: Example of using storage functionality to perform distributed placement of data on virtualized disks

This example groups four RAID groups for DB area files together to form one storage pool (virtual disk) and uses a striping method that distributes data to the four RAID groups. Several logical units are created by using virtual disks and are allocated as block special files that make up the DB area files.

When data is accessed in the DB area files, the I/O operations are distributed almost evenly among the four RAID groups. Such a configuration enables you to issue more I/O requests than when there is only one RAID group. This example separates the storage pool for DB area files from other storage pools.

However, if there are not enough available RAID groups, use a single storage pool to apply priority to I/O operations on the DB area files.

For details about how to configure storage, see the documentation for the storage product.

■ Example 2

The following figure shows an example of a striping configuration that uses a logical volume manager (LVM) supported by the OS.

Figure 6‒2: Example of a striping configuration that uses a logical volume manager (LVM) supported by the OS

This example registers four RAID groups for DB area files into one volume group and creates logical volumes from this volume group with a striping specification. When data is accessed in the DB area files, the I/O operations are distributed almost evenly among the four RAID groups. Such a configuration enables you to issue more I/O requests than when there is only one RAID group. This example separates the volume group for DB area files from other volume groups.

However, if there are not enough available RAID groups, use a single volume group to apply priority to I/O operations on the DB area files.

For details about how to configure LVMs, see the OS documentation.

To Page Top