Using the ResUsageSAWT table to Monitor AMP Worker Tasks - comment by carrie

Hi Helena,
 
If you are running out of AMP worker tasks (AWT) that can slow a query down.   Query steps require AWTs to execute, and if all the AMPs a query step is running in do not have an available AWT at the time message containing that step arrives, then that query step will have to wait.  This can cause queries to take longer.   If you are running out of AWTs at the time that you are experiencing query performance problems, that could be a contributing factor.  
 
There are several factors that can contribute to higher than usual AWT counts.  Some of them include:
 
1.   You have more user work on the platform than you have AWTs to support (high work00 and work01 counts)
2.   Many rollbacks are running, using up AWTs (high Work12 and Work13 counts)
3.  One or more queries have skewed processing and are holding onto AWTs on some AMPs a lot longer than on others AMPs (high work00 and work01 counts but only on one or a few AMPS)
4.  You have a "penalty box" workload with a low cap on CPU usage where a large number of queries have been placed, all of which are holding onto 1 or more AWTs (high work00 and work01 counts)
 
You can find out more about monitoring AWTs by looking at the AMP Worker Task orange book.  Orange books are available at Teradata At Your Service, or you can ask your Teradata account team for assistance in finding that orange book. 
 
Thanks, -Carrie