Variable UB-Trees: An efficient way to accelerate OLAP queries1

Pre-computation, clustering and indexing are common techniques to speed up query processing. Precomputation results in the best query response time at the expense of load performance and secondary storage space. For data warehousing (DW) applications, pre-computation is mostly discussed for aggregation operations [CD97]. Indexing is used to efficiently process a query if the result set defined by the query restrictions is fairly small. Favoring retrieval response time over update response time allows to build several indexes on one table or data cube of a data warehouse. Bitmap indexes are widely discussed as an improvement over B-Trees for DW applications, since they efficiently evaluate queries with multi-attribute restrictions. However, the overall result set still must be relatively small. This is a major drawback of bitmap indexes, since usually a relatively large part of a cube must be accessed in order to calculate aggregated measures. Clustering places data that is likely to be accessed together physically close to each other. The goal of clustering is to limit the number of disk accesses required to process a query by increasing the likelihood that query results have already been cached.