Optimal Checkpoint Times for a Majority Database System | ||||
Author | Kenichiro Naruse
|
|||
Co-Author(s) | Toshio Nakagawa
|
|||
Abstract | The tasks with random processing times are executed successively. Two types of checkpoints can be placed at the end of tasks. We suppose several database-nodes and one management-node. Tasks are executed by the database-nodes and watch them by management-node. We consider two types of nodes that are single node and majority node. Two types of checkpoints are called Journal Checkpoint (JC) and Flush Checkpoint (FC). JC needs a small time compared with FC to store checkpoint data, and FC needs a long time compared with JC to store checkpoint data. To get quickly response from the system, we divide database data by the number of database-nodes to store. The problem is that in what places we should set suitable checkpoints and how many database-nodes we should prepare. We derive how many number of JC checkpoints between FC to minimize the execute time, and how many number of database-nodes to minimize it. The mean execution times per one task are obtained and optimal numbers that minimize them are derived analytically and numerically.
|
|||
Keywords | Reliability, Checkpoint times, Big data, Database system, Majority systems, High speed processing | |||
Article #: 19153 |
August 5-7, 2013 - Honolulu, Hawaii, U.S.A. |