Two-step backup mechanism for real-time main memory database recovery

Proposes a recovery mechanism for real-time main-memory database systems (RT-MMDBSs) which aims at reducing the system recovery time, as well as ensuring a fast response time and predictability. Conventional main-memory database (MMDB) recovery schemes usually use logging with checkpointing, which involves many disk I/Os during transaction processing. Consequently, it not only ensures the predictability of the response time, but also fast recovery, because undo/redo logs should be analyzed and applied after failure. Our recovery mechanism replaces logging and checkpointing activities with a single backup activity, which results in eliminating the need for log writing. Furthermore, it performs transaction processing and backup processing in separate areas, as a result minimizing their interference with each other. We use part of the main memory as a shadow memory area and the stable memory as a backup buffer. In our scheme, a two-step page-based backup is made: the first backup step is done from the shadow memory to the backup buffer, and the second step is performed from the backup buffer to a backup disk. We briefly show the worst-case performance analysis of our mechanism