Managing Analysis Services Partitions
A partition is a file on a hard disk that contains a subset of the data included in an Analysis Services database. Partitions let you spread data over multiple hard disks. This includes combinations of both local (stored locally on hard disk) and remote (distributed across multiple hard disks) partitions. Partitions rely on storage settings to define the format and processing schedule for the database, and they use writeback settings to enable what-if analysis. What-if analysis enables a user to input their own data and evaluate the changes that cascade throughout their cube. For more information about storage settings, see Configuring Storage. For more information about writeback settings, see Setting Partition Writeback and Write-Enabled Partitions.
Local partitions are partitions that are defined, processed, and stored on one server. If you have large measure groups in a cube, you might want to partition them out so that processing occurs in parallel across the partitions. The advantage is that parallel processing provides faster execution. Because one partition processing job does not have to finish before another starts, they can run in parallel. For more information, see Creating and Managing Local Partitions. For implementation of storage design for a cube, by using the Partitions tab in Cube Designer, see Designing Partition Storage and Aggregations for more information.
Remote partitions are partitions that are defined on one server, but are processed and stored on another. If you want to distribute storage of your data and metadata across multiple servers, use remote partitions. Ordinarily, when you transition from development to production, the size of data under analysis grows several times over. With such large chunks of data, one possible alternative is to distribute that data over multiple computers. This is not just because one computer cannot hold all the data, but because you will want more than one computer processing the data in parallel.
If two partitions use the same aggregation design, you can merge those two partitions into one. For example, if you have an inventory dimension that is partitioned by month, then at the end of each calendar month, you can merge that month partition with the existing year-to-date partition. This way, the current month partition can be processed and analyzed quickly, while the rest of the year in months only has to be reprocessed when merged. That reprocess requires longer processing time, and can be run less frequently. For more information about managing the partition merging process, see Merging Analysis Services Partitions. To edit cube partitions by using the Partitions tab in Cube Designer, see Editing Partitions.