Inside OLAP Engine: Cache Prefetching (Mosha Pasumansky)

I am writing a whitepaper about Analysis Services 2005 Cache Manager, and it gets longer and longer. I already spent few days on it, and I am still not even halfway done. So I decided to publish an excerpt from this whitepaper (chapter 2.1.5) and gauge whether I am getting it at the right technical level or whether I am going too deep or too shallow. The timely comments will be appreciated.

As we learned by now, the FE query optimizer decomposes the MDX query into multiple single granularity subcubes. Depending on which cell or group of cells are being computed at the moment, FE will choose the appropriate subcube to send request for data to SE. However, since the query optimizer has the global picture of the entire query, it can tell which cells will be requested in the future. This knowledge is used by the Cache Manager to reorder the subcube requests in order to take advantage of datacache filtering and aggregation described in the previous chapter. This is called cache prefetching Cache prefetching is best illustrated by example.

Read more...