Budget Variance - A study of MDX optimizations: evaluation modes and NON_EMPTY_BEHAVIOR
According to OLAP Survey, the number one objective in building OLAP systems is performance. Certainly postings in newsgroups and forums confirm that finding. It is safe to say, that the most frequent question people are asking is "How do I optimize the following MDX...". Optimizing MDX is both science and art. It requires a mix of knowledge of MDX constructs, familiarity with UDM concepts and some basic understanding of how query optimizer works. But what I find the most difficult, is to understand the real problem behind the question. Most of those require deep insight into the concrete model and intimate knowledge of the business logic driving the requirements. Often these details obscure the technical question enough that either nobody is brave enough to take a shot at it (certainly, myself, if I didn't understand the issue after reading it - I am unlikely to engage) or a long thread starts which sinks in additional details and clarifications, and at the end even if the answer is both correct and improves performance, it is not easy to extract the core of the idea behind the optimization and generalize it enough to apply to other scenarios.