On the setting changes, stats, as stated will help with counts as that info is precalculates and stored in the metadata. The CBO and stats also help a lot with joins. It is possible that the OS cache is more to do with the improvement if this was a subsequent run with little activity. You could look at Hive on Spark for better consistent performance. Set hive.execution.engine = spark; On the times, the big impact between job submission and start is the the scheduler. That is a deep topic. It is best if you read up on them and review your settings and ask any specific questions that come up, preferably in a new topic. The other factor, not captured on the job stats, is the time it takes to return the results to the client. This will vary depending on the client and there isn't much to do about it. In general small result sets can be handle by the hive CLI. You can increase the client heap if needed. Otherwise use HS2 connections like beeline or HUE.
... View more
