Archives of the TeradataForum

Message Posted: Fri, 14 Oct 2005 @ 09:48:19 GMT


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

Subj:		Re: NUSI Access Performance Issues

From:		Dunweber, Ole

Hi Clay,

you wrote:

(in our case, an AJI.) That means it can't have anything to do with PPI (we also see bad NUSI performance, but that's with PPI-tables, and NUSIs are not "PPI-aware" before release 6.0)

In almost all of these cases, the optimizer's estimates are way off*, and the use of a NUSI is a very poor decision. So, we ran a test using the same queries against a fact table without NUSIs.

I guess it's the same fact-table (AJI?)

Is it the estimated time that are way off, or the estimated rows?

If it's the estimated rows, then maybe the optimizer is not costing the use of AJI's correctly?

As I understood at Partners the decision to use JIs are in some cases "heuristic"?

Is there anything specific to the NUSI access method which makes it particularly susceptible to performance problems on a busy system?

Maybe this makes some sense:

I guess NUSI-access tend to be more skewed than other accesses (if the table is well distributed).

And if you use milestones in Priority Scheduler, then the skewed AMPs would reach the milestone(s), and therefore be demoted to a lower priority, making it run even longer.

Ole


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

Attachments

Library

Quick Reference