• Summary
• Related Documents
  • Active Bibliography
  • Co-citation
• Version History

Abstract:

With today's demands for continuous availability of mission-critical databases, on-line reorganization is a necessity. When records are moved during an on-line reorganization of a database, references to the moved records must be changed to refer to the new location. In particular, secondary indexes must be updated. In this paper we show how to update secondary indexes efficiently when moving records from an RID organization to a primary B+tree. Our new on-line reorganization algorithm defers secondary index updates and piggybacks them with user transactions, significantly reducing the total I/O cost. We carefully detail the concurrency (locking) and recovery (logging) necessary to assure that almost all the database is available all of the time and that the reorganization is interruptible. Our method uses a page-oriented approach. Whenever a secondary index leaf page is brought into the database buffer by a user transaction, we check small main-memory-resident forwarding address tables to see if there are any pending updates on that index leaf page. When the forwarding address tables are full, a clean-up process temporarily takes over from the reorganization process and brings the necessary secondary index leaf pages into the buffer for update. This will free space for new forwarding addresses. Both an analysis and a performance study are done to measure the savings in I/O over other methods. Even though we discuss our algorithms in the context of a particular on-line database reorganization, we believe our method can be used as a general technique for deferred secondary index updates. It is efficient in I/Os, interruptible without loss of work, and provides high availability through minimal locking.

Citations