Re: Periodic Stats Collection -- CBO Stats Myth? -- delta refresh of statistics

From: jaromir nemec <>
Date: Mon, 6 Oct 2008 21:15:05 +0200
Message-ID: <513901c927e7$d77d6de0$3d02a8c0@ADLA>

Hello Tim, Wolfgang,

> Just how inaccurate is part of the game-theory that we all use when deciding between COMPUTE STATISTICS and ESTIMATE STATISTICS.

I'm wondering why Oracle doesn't support the third option: delta refresh of the statistics. With the strong support of materialized view and fast refresh, I can imagine this will be a very similar concept. Currently if I use the terms of MV only full refresh of statistics is possible (or a full refresh based on sample).

> .and determine the min and max values for columns based on the values in the rows returned by the sample.

The min and max will be solved natively. Possible, some obstacles will arise by distinct values and histograms. But if I understand it correctly, there already exist some algorithm solving this for maintaining the global statistics of a partition table after gathering of the statistics of a single partition. (I'm not sure if this algorithm is published). Of course there will be data categories where the delta refresh would not work very precise, but I guess for majority of business data it could yield good results.

Any comments?


Jaromir D.B. Nemec

Received on Mon Oct 06 2008 - 14:15:05 CDT

Original text of this message