E. Omiecinski, L. Lee, and P. Scheuermann. Concurrent file reorganization for record clustering: A performance study. In Proc. of 8th International Conference on Data Engineering, pages 265--272, 1992.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....and major DBMS providers. 18] investigated how to compact a primary B tree which had become sparse and and [19] described a method for defering secondary index updates. 8] examined the problems of changing from one access method to another, such as from B tree to linear hashing and [9] studied how indexed sequential files can be compacted and clustering organizations described by hypergraphs can be created. 14] detailed methods to increase concurrency during the restoration of clustering indexes in IBM s DB2. 7] described restartable algorithms for online construction of an ....

E. Omiecinski, L. Lee, and P. Scheuermann. Concurrent file reorganization for record clustering: A performance study. In Proc. of 8th International Conference on Data Engineering, pages 265--272, 1992.

....this reason, there has been a resurgence of interest in on line incremental reorganization algorithms. Papers on on line construction of secondary B tree indexes [Mohan1992a, Srinivasan1991, Srinivasan1992] conversion from B trees to hashing [Omiecinski1988] reclustering of records [Omiecinski1992], resequencing of primary B tree leaves [Smith1990] and use of partial indexes[Stonebraker1989] have recently appeared. In this paper, we consider only one aspect of the on line reorganization problem, an aspect which has been ignored in the other papers. This is the changing of all ....

....the length of the batch transaction is a parameter which can be tuned to system requirements. We do not discuss batch size further; in the remainder of the paper, we assume only one record is moved per transaction. The extension to batching is straightforward. An alternative approach, as in [Omiecinski1992], would be to make a differential table or look aside table correlating new and old addresses after a record is moved. We believe there is little advantage to postponing the updating of references and maintaining a differential table. While there are still entries in the table, every database ....

E. Omiecinski, L. Lee and P. Scheuermann. Concurrent File Reorganization for Record Clustering: A Performance Study. Eighth International Conference on Data Engineering, 1992. pp. 265-272.

....keeping the database available Clustering is not the only reason to migrate from one structure to another. Changes of database schema and changes of user query patterns, etc. can all lead to database reorganization. Although there has been a lot of research on on line database reorganization [37, 28, 29, 30, 31, 34], the results have some unsatisfactory or incomplete aspects. For example, it is shown in [34] how to do the conversion between two physical storage structures correctly by encapsulating all the reference changes in one transaction, but it is still not clear how to do it efficiently. On line ....

....a portion of the database as possible at one time to avoid interfering user transactions. In addition, an important contribution of the thesis is to integrate on line reorganization with standard concurrency and recovery algorithms used in large modern DBMSs. Much of the research in the past [29, 30, 31] does not take this into account, making it unusable for practitioners. The reorganization algorithms in this thesis are carefully designed, and can be easily integrated into existing database systems. Efficient Reference Updating Algorithm The reference updating algorithm makes the updates ....

[Article contains additional citation context not shown here]

E. Omiecinski, L. Lee, and P. Scheuermann. Concurrent file reorganization for record clustering: A performance study. In International Conference On Data Engineering, pages 265--272, 1992.

....all the necessary changes relating to moving one record into one transaction. In [Omi88] records are moved from a primary B tree to a linear hashing organization, one leaf at a time. The same space is used. This is an in place reorganization. Logging and recovery is not discussed. In [OLS92], records are reclustered in place. Logical pages are sets of records one wishes to place in one physical page. One by one, all the logical pages are transformed to physical pages by bringing all their records into the buffer and moving them. References to moved records are changed later. A list ....

E. Omiecinski, L. Lee, and P. Scheuermann. Concurrent file reorganization for record clustering: A performance study. In International Conference On Data Engineering, pages 265--272, 1992.

....may change so that clustered data and access patterns are no longer nicely matched. A static clustering scheme takes advantage of one layout of the data exclusively. Reclustering is not performed. Dynamic clustering tries to recluster data depending on the current performance of the system [4, 8 10, 17, 22 27, 37]. Typically, performance depends on the current application s access patterns. Dynamic clustering is categorized as on line and off line. Off line clustering means that the system is periodically restricted from user access while the entire database is reorganized. The major challenge of ....

Omiecinski, E. and Lee,E., "Concurrent File Reorganization for Record Clustering," Proceedings of the Eighth Conference on Data Engineering, 1992. Concurrent reorganization is a form of incremental reorganization that uses multiple-processor architectures to recluster objects. An algorithm is given.

....efficient ways of reusing the number after deletion, and test the method using a mixture of updates and queries. Other future plans include extending this work to shared hierarchies, where children have more than one parent. We would also like to work on reorganization [SC91] MN92] SD92] [OLS92] so that we may specify how to take some other organization and transform it to an Enc organization, or the reverse, without taking the database off line or making a full copy. Other related problems like integration the Enc method with existing database utilities would also be interesting. ....

E. Omiecinski, L. Lee, and P. Scheuermann. Concurrent file reorganization for record clustering: A performance study. In International Conference On Data Engineering, pages 265--272, 1992.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC