摘要 |
Provided is a database management system capable of executing shared nothing parallel searches using a decentralized storage group comprising a plurality of site partition groups. Therein, operation which maximizes operating efficiency is achieved without generating defects in shared nothing parallel search results. Specifically, a plurality of compensation processing methods are selected to handle a situation in which an error has occurred during search processing so that defects do not occur in shared nothing parallel search results. In addition, an expected value for the overall processing cost, which takes into account compensation processing and is based on the operating efficiency and failure rate calculated from the performance statistics data and reliability data pertaining to the decentralized storage group, is calculated using each of said compensation methods, and the query division of parallel searches is stipulated on the basis of compensation processing with a small expected value for the overall processing cost. |