As shown in the recipe steps, the issue was due to the result of carrying an unknown dimension.
Randomly distributing the unknown dimension on a range of negative key values is an effective means to address this problem.
80% of the business is done by a small number of customers. Primary indexes might be fine, but you still have data skew on dynamic joins. Reducing skewness on joins is a major challenge for developers when designing the process, be it ETL or business reports. Skewed joins may over kill the system and can help in arising hot AMP situations on the Teradata box.
Sub-optimal or skewed queries can be identified at runtime by looking at viewpoint, which will have a skew percentage displayed on the session information ...
