Rapid Recovery of Very Large Memory Resident Data Objects

Advancements in technology have enabled many computer vendors to offer computer systems with gigabytes of memory at affordable prices. The availability of an abundance of real memory in a computer creates new opportunities for software designers — very large data objects can be made memory resident to support very efficient processing. However, crash recovery becomes more time consuming because of the large amount of memory resident information that need to be saved and restored. In this paper, we propose an efficient recovery scheme based on mapping very large memory resident data objects to files. The underlying memory mapped file mechanism will propagate updates on the data objects to the mapped files automatically and checkpointing these data objects becomes a simple matter of forcing the operating system to write out all dirty buffers associated with them. Changes to data objects in between checkpoints are logged. Recovery is accomplished by restoring the most recent checkpoint and then redoing all the changes logged since this checkpoint to bring the system back to where it is immediately before the crash.