Reliability is the ability of an item to perform a required function under stated conditions for a stated period of time.

Buffer primitives

The main memory acts as a buffer from the disk storage, and supports a set of primitives called by providing a page and used for moving it in and out of the disk.

• Fix: searches the page in the buffer, if already there increment usage counter and return reference, otherwise selects a free page in the buffer in which to put it. If the page was dirty, it flushes it.
• Unfix: de-allocates page from the buffer and decrements usage count.
• Force: transfers synchronously a page from buffer to disk.
• Set dirty: modifies page status to denote that it has been changed.
• Flush: transfers asynchronously pages from buffer to disk, when page no longer needed.

Normally, a buffer has as default settings “no steal” and “no force”, meaning that when all pages are taken, the new transaction must wait until one frees, instead of force-unloading one, and pages are not transferred right after commit, but the manager can postpone the task.

Logs

Log management rules ensure transactions implement write operation in a way that reliability is supported. A log record must be written synchronously, storing the state modified by the write both before and after the commit. If failure is detected, uncompleted transactions require that the state is taken back to the pre-commit log, while unordered ones require both undos and redos operations.