Archives of the TeradataForum

Message Posted: Tue, 08 Nov 2005 @ 21:36:05 GMT

	<Prev	Next>		<<First	<Prev	Next>	Last>>

Clay,

Parsing Engine activity can be slowed down by some of the same resource bottlenecks as AMP activity. If you are running canary queries on your AMP side, it would be interesting to see if longer-than-normal PE times can be correlated to longer-than-normal canary query times.

Both statement and dictionary cache are flushed at the same time, so when that happens on a PE, more parsing will take place at the same time as there is more effort to access dictionary information that might previously have been cached. If you have to do one or more IOs for dictionary information and IO is your bottleneck at that point in time, there could be a wait involved in performing the IO that is needed for the parsing.

If you are memory constrained it is possible that a part of memory needed to use a cached plan has been swapped out, and there will be an extra effort to bring that back in.

Also, PE CPU is prioritized the same as AMP CPU (based on the performance group that the user is associated to), so it is possible there is a priority issue involved. If there are lots of resource-intensive queries in the same allocation group on that node fighting over a small entitlement of CPU, needs of the parsing engine tasks are not protected. If the user of the query has a low priority, AND that allocation group has high concurrency, AND the query needs parsing/optimization that is non-trivial, AND you are memory constrained at that time, AND your dictionary cache on that PE was just flushed, all those conditions could contribute to a longer time spent in the PE.

Thanks, -Carrie

	<Prev	Next>		<<First	<Prev	Next>	Last>>