Multidimensional aggregation is a dominant operation on data ware-houses for on-line analytical processing (OLAP). Many efficient algorithms to compute multidimensional aggregation on relational database based data wa...Multidimensional aggregation is a dominant operation on data ware-houses for on-line analytical processing (OLAP). Many efficient algorithms to compute multidimensional aggregation on relational database based data warehouseshave been developed. However, to our knowledge, there is nothing to date in theliterature about aggregation algorithms on multidimensional data warehouses thatstore datasets in multidimensional arrays rather than in tables. This paper presentsa set of multidimensional aggregation algorithms on very large and compressed mul-tidimensional data warehouses. These algorithms operate directly on compresseddatasets in multidimensional data warehouses without the need to first decompressthem. They are applicable to a variety of data compression methods. The algorithmshave differefit performance behavior as a function of dataset parameters, sizes of out-puts and main memory availability. The algorithms are described and analyzed withrespect to the I/O and CPU costs. A decision procedure to select the most efficientalgorithm, given an aggregation request, is also proposed. The analytical and ex-perimental results show that the algorithms are more efficient than the traditionalaggregation algorithms.展开更多
文摘Multidimensional aggregation is a dominant operation on data ware-houses for on-line analytical processing (OLAP). Many efficient algorithms to compute multidimensional aggregation on relational database based data warehouseshave been developed. However, to our knowledge, there is nothing to date in theliterature about aggregation algorithms on multidimensional data warehouses thatstore datasets in multidimensional arrays rather than in tables. This paper presentsa set of multidimensional aggregation algorithms on very large and compressed mul-tidimensional data warehouses. These algorithms operate directly on compresseddatasets in multidimensional data warehouses without the need to first decompressthem. They are applicable to a variety of data compression methods. The algorithmshave differefit performance behavior as a function of dataset parameters, sizes of out-puts and main memory availability. The algorithms are described and analyzed withrespect to the I/O and CPU costs. A decision procedure to select the most efficientalgorithm, given an aggregation request, is also proposed. The analytical and ex-perimental results show that the algorithms are more efficient than the traditionalaggregation algorithms.