There is a lot of resources about Microsoft Analysis Services. There are literaly hundrends of Web pages available (see http://www.mosha.com/msolap), there are dozens of books written (http://www.mosha.com/msolap/books.htm) etc. They cover all the possible subjects about Analysis Services but one - how does the engine works internally. Such information is very scarce and difficult to find. We, as a product team, occasionally give presentations about Analysis Services internals, but the content of those presentations isn't usually publicaly available. So I was surprised to discover several presentations from TechEd 2002 on www.only4gurus.com, and one of them is called “Analysis Services: Server Internals”. This presentation was originally prepared by my old-time friend and colleague - Sasha (Alexander) Berger together with Richard Tkachuk, and later used by Tom Conlon. To understand the context of the presentation, let's take a look at one of the first slides - supersimplified architecture of Analysis Services engine:
Sasha Berger is an architect and development lead for the Storage Engine component, so naturally presentation deals with the such storage engine concepts as aggregations, on-disk record layout, indexing schemes, dimensions structure, processing, memory management, partitions, distinct count etc, but it doesn't talk about MDX, query optimizer and other aspects of Formula Engine component (the one that I am responsible for). Lastly, the relevance of this presentation to Analysis Services 2005 varies. Dimensions architecture was greatly changed in Yukon, to account for attribute based UDM, but also to enable even more scalable dimension architecture by not requiring the entire dimension to be loaded into server memory. But partitions and indexing still have a lot of in common with Analysis Services 2000, and many slides which talk about map etc would still be applicable.