Deferred Maintenance of Indexes and of Materialized Views
Maintenance of secondary indexes and materialized views can cause the latency and bandwidth of concurrent information capture to degrade by orders of magnitude. In order to preserve performance during temporary bursts of update activity, e.g., during load operations, many systems therefore support deferred maintenance, at least for materialized views. However, deferring maintenance means that index or view contents may become out-of-date. In such cases, a seemingly benign choice among alternative query execution plans affects whether query results represent the latest database contents. We propose here a system that distinguishes between the maintenance of of logical contents and physical structure. This distinction lets us compensate for deferred logical maintenance operations while minimizing the impact of deferred physical maintenance operations, and results in support for concurrent high update rates and immediate, index-based query processing with correct transaction semantics.
KeywordsQuery Result Query Execution Partial Index Primary Index Load Operation
Unable to display preview. Download preview PDF.
- 2.Bender, M.A., Farach-Colton, M., Mosteiro, M.: Insertion sort is O(n log n). In: Fun with Algorithms, pp. 16–23 (2004)Google Scholar
- 4.Graefe, G.: Sorting and indexing with partitioned B-trees. In: CIDR (2003)Google Scholar
- 5.Graefe, G.: Implementing sorting in database systems. ACM Comput. Surv. 38 (September 2006)Google Scholar
- 6.Graefe, G.: A survey of b-tree locking techniques. ACM Trans. Database Syst. 35, 16:1–16:26 (2010)Google Scholar
- 8.Graefe, G., Kuno, H.: Self-selecting, self-tuning, incrementally optimized indexes. In: Proceedings of the 13th International Conference on Extending Database Technology, EDBT 2010, pp. 371–381. ACM, New York (2010)Google Scholar
- 10.Hamilton, J.: Spot instances, big clusters, & the cloud at work. Amazon Cloud Blog Post (September 2011)Google Scholar