分散システムにおける意味的な復旧方法

概要

論文の詳細を見る
In distributed applications like teleconferences, a group of multiple autonomous objects are required to be cooperated by sending messages through communication networks to achieve some objectives. If an object o is faulty, o and objects which have received messages sent by o have to be rolled back to the checkpoint by restoring the information stored in the log taken at the checkpoint and then the computation on o is restarted [1, 2]. Leong and Agrawal[2] present the concept of significant messages if the state of an object is changed on receipt of a message m. If o is rolled back, only objects which have received significant messages sent by o are rolled back. In the distributed computation, objects send kinds of messages, i. e. request, response, and data messages. In the significant messages, the transmissions of the request, response, and data messages are not considered. In this paper, we would like to define influential messages by taking into account the kinds of messages sent by the objects. Then, we would like to discuss object-based checkpoints which can be taken from the object point of view while it may not be consistent. In section 2, we first present the system model. In section 3 we discuss the influential messages and define the object-based checkpoint. In section 4, we showhow the number of checkpoints can be reduced by the object-based checkpoint.
一般社団法人情報処理学会の論文
1996-09-04