非同期分散システムにおけるチェックポイントとロールパック

概要

論文の詳細を見る
Distributed applications are realized by cooperation of multiple processes executed in multiple computers. These processes are not always reliable and available. Checkpointing and rollback are well-known time-redundant techniques in order to allow processes to make progress even if some processes fail. The processes tike checkpoints by saving their state information in the local logs while executing applications. If the processe8 fail in the system, the processes are rolled back to the checkpoints by restoring the saved state information and then restarted from the checkpoints. In the conventional methods, all the processes are synchronized by using such a protocol as the two-phase commitment protocol [1]. In this paper, we would like to discuss a new method where the processes are allowed to be asynchronously rolled back and restarted.
一般社団法人情報処理学会の論文
1996-09-04