Memory Allocation

It is important to provide sufficient physical memory, on a server dedicated to building cubes, to avoid excessive pagination. In addition to having sufficient memory to handle all running application requests, make sure the operating system disk cache on the server can grow as large as required during the cube build process.

Once you have optimized the memory settings for your server, track virtual and working memory use to find any bottlenecks created during the processing of categories in your largest slowest cubes.

Typically, total addressable (virtual) memory usage climbs rapidly during category generation, and remains relatively constant during all cube build phases (read data, update metadata, and update cube). Working memory, or that portion of the physical memory that is used by Cognos® Transformer, also rises quickly and remains high until the cube update phase, when it is freed for use by the operating system disk cache.

In the first phase, the more categories there are, the more memory is required. Although models differ, working memory usage is typically 500-1,500 bytes per category. Paging (swap file use) occurs if the space allocated to working memory is insufficient and cannot grow to the amount required to process the model.