Cube aggregation data is stored in one or more partitions. Each partition can be stored in one of three modes: MOLAP (multidimensional OLAP), ROLAP (relational OLAP), or HOLAP (hybrid OLAP). These storage modes interact with the data warehouse in different ways.
You can explicitly define a filter in a partition’s definition that specifies the data applicable to the partition. Explicit partition filters minimize unexpected effects on OLAP data that can be caused by changes to the underlying data warehouse data.
MOLAP partitions completely contain all assigned data, including precalculated aggregations and facts from the data warehouse fact table. Because all queries to a MOLAP partition can be answered from data stored in the partition, changes to data warehouse data or structures do not affect the partition until the partition is processed to rebuild its structure or update its data. You can use an explicitly defined filter to restrict the data that will be read from the data warehouse when the partition is processed.
ROLAP partitions store precalculated aggregations in tables automatically created in the data warehouse database. Some queries to ROLAP partitions require reading data from the fact table, so changes to the fact table can affect the answers to these queries. Answers to some queries can require combining precalculated data stored in aggregation tables with specific facts from the fact table, or with calculations performed on facts in the fact table. If data in the fact table is changed, inaccurate answers can result until the partition is processed to recalculate the aggregations.
You can use an explicitly defined partition filter to specify which data will be read from the data warehouse. Data added to the data warehouse that does not match the filter criteria is ignored by queries to the partition until the partition’s definition is changed to include the new data. However, changes to data that existed in the data warehouse when the partition was processed can cause the partition to provide incorrect answers to queries until the partition is processed again.
HOLAP partitions combine characteristics of MOLAP and ROLAP partitions. Precalculated aggregations are stored in a multidimensional structure that is like that of a MOLAP partition except that the fact table is not copied into the structure. Changes to the data warehouse fact table affect HOLAP partitions the same way they do ROLAP partitions. Queries to the partition that can be answered using only precalculated information in the multidimensional structure are unaffected by changes to the fact table data, but queries that access the fact table can return incorrect answers if the queries also access precalculated information from the multidimensional structure.
As with ROLAP partitions, you can use an explicitly defined filter in a HOLAP partition’s definition to specify which data will be read from the data warehouse.