Using Raw Device Partitions
The term 'raw devices' applies to the character oriented disk device files (as opposed to the block oriented ones) normally found in /dev. These device files are a part of the interface between the hardware disks and the UNIX system software.
In most UNIX systems it is a performance advantage to use raw device files for data storage. By using raw devices, the UNIX file system (which uses index pages to locate the file blocks) is bypassed and the operating system is able to perform more effective I/O.
Of course, the effect also depends on several other factors such as file size, transaction rate, I/O-implementation, etc. Generally, large and/or frequently used databanks should be stored in raw device files.
Note: Familiarity with raw device files is recommended. It is essential that a raw device file be correctly defined, since the normal UNIX file handling safeguards do not apply.
Some important points to be considered:
- Usually the first cylinder on the disk (cylinder 0), or corresponding, should be avoided when partitioning, since operating system information for the disk is stored there.
- Overlapping partitions must be manually avoided.
- It should be verified that backup procedures include raw devices.
- Always take a complete backup before enabling use of a raw device.
The major disadvantage of using a raw device file is that the maximum file size is fixed by the size of the partition. If the partition becomes full, the raw device file must be moved to a larger partition, if one is available.
In the worst case, the disk must be reformatted in order to create a larger partition.
Use of the raw device interface is completely transparent from the point of view of a Mimer SQL user. A user has no way of distinguishing a system using raw device files from a system where ordinary UNIX files are used to store databanks.
Upright Database Technology AB
Voice: +46 18 780 92 00
Fax: +46 18 780 92 40