An efficient recovery scheme for locking-based distributed database systems

This paper presents a checkpointing recovery scheme which effectively copes with media failures in distributed database systems, under the two-phase locking policy. Our scheme utilizes both the current database area and the checkpoint area, so that the transactions whose effects are to be included in the checkpoint and the ones whose effects are to be excluded from the checkpoint can separately reflect their effects. Hence, the check pointing process under our scheme is executed concurrently with normal transactions, without aborting or delaying any of them. Unlike the existing schemes, each data item needs to be read at most once for checkpointing without the maintenance of the before-images of the data items