Improved UFS Direct I/O Concurrency

The performance of direct I/O, which is used by database applications to access unbuffered file system data, has been improved by allowing concurrent read and write access to regular UFS files. Previously, an operation that updated file data would lock out all other read or write accesses until the update operation was completed.

Concurrent writes are restricted to the special case of file rewrites. If the file is being extended, writing is single threaded as before. Generally, databases pre-allocate files and seldomly extend them thereafter. Therefore, the effects of this enhancement are evident during normal database operations.

The direct I/O improvements brings I/O bound database performance on a UFS file system to about 90% of raw partition access speeds. If the database is CPU bound or bus bandwidth bound, there might be no improvement.

Consider running your I/O database applications with direct I/O enabled if you are already using UFS to store database tables. Use your database administrative procedures to enable direct I/O, if possible. If there is no way to enable direct I/O through your database product, use the mount -forcedirectio option to enable direct I/O for each file system. Or, use the directio(3C) library call to enable direct I/O.

See mount_ufs(1M) or directio(3C) for more information.