Archives of the TeradataForum

Message Posted: Wed, 22 Jun 2011 @ 22:24:51 GMT

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

<-- Anonymously Posted: Wednesday, June 22, 2011 18:16 -->

Tomas,

The PE Parsing Engine actually has two memory caches, one is the Request Cache, which is used to cache execution plans and so speed up repetitive, identical SQL requests, like tactical ones, the second is the Dictionary Cache, which is used to speed up for instance access to DDL definitions for the objects, avoiding repetitive reads of same info stored in the DD/D.

To spoil or drop the table DDL in the PE Dictionary Cache is always due because we run something which makes a change to its DD/D info within same query, it also occurs as a housekeeping function where the PE spoils the info every 4 hours, but it is only the first case the one we could ever see as a step of the execution plan in the EXPLAIN report.

So it is not logical to have such a step like #4 in your first EXPLAIN if we are just running a DML like your global, all-AMPs DELETE, and not for instance a DDL for the table. It looks like the information provided has been edited, at least for database, table and column names.

Anyway this #4 step would never harm the overall performance of your execution plan, it may take a few microseconds, and it ocuurs once your massive DELETE is over.

Also this is not the kind of execution plan that would be affected by flushing the Request Cache, so your statement about having a significant slow-down for your DELETE is not true.

Hope this helps.

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