Checkpointing Schemes for Fast Restart in Main Memory

The potential for substantial performance improvement in a main memory database system (MMDB) is promising, since I/O activity is kept at minimum. On the other hand, due to the volatility of main memory, the issue of failure recovery becomes more complex than in traditional disk resident database systems. In this paper, we present four checkpointing schemes for the MMDB. The proposed schemes are novel in the sense that they can reduce the amount of disk I/O by exploiting recovery memory as a checkpointing buffer. Furthermore, logging overhead is reduced by maintaining only REDO logs in the stable memory. We evaluate the performance of the proposed schemes under a variety of database workloads using a centralized database simulation model.