Wednesday, September 30, 2015

Comparison of a Stored and Dynamic ASO Hierarchy

I've heard from many people on many occasions that when designing an ASO database you should make as many dimensions Stored as possible. Just how much will this improve performance?

There are many different way this can be tested. I will start out with a simple test case to gauge retrieval performance. I have four scenarios in mind for use with the ASOsamp database:

1. Store dimension set to Stored. No aggregations.
2. Store dimension set to Dynamic. No aggregations.
3. Store dimension set to Stored. Recommended aggregations.
4. Store dimension set to Dynamic. Recommended aggregations.

I will run two different queries to measure retrieval performance. Query 1 will be the default retrieve. Query 2 will zoom to all levels of the Store dimension with the following grid (all twelve months are not shown for size purposes).

The Results

The Analysis

For Query 1 with no aggregations, the Stored dimension is faster but on Query 2, the Dynamic dimension is faster. With aggregations, Query 1 is the same between Stored and Dynamic. With aggregations, Query 2 is slightly faster with the Dynamic dimension than with Stored.

I have no idea why the Dynamic dimension is faster for Query 2. Any reason I could come up with would be purely speculative. Is it the size of the input data, the distribution of the data, the size of the dimension, the full moon? I have no clue. I did rerun these tests many times so it wasn't just a fluke -- the results were consistent. I'm going to have to explore this further with more testing. Any insights or ideas on further test cases on this would be appreciated.

No comments: