The new Summary Collect Statistics feature starting with Teradata Release 14

Diagnostic helpstats on for session; （先执行这句话，然后再explain的话会提示你要添加的collect stas）

explain

select * from talbe

Summary statistics are a very useful feature, which is new in Teradata Release 14. Contrary to traditional statistics, which are collected on columns or indexes, summary statistics are table level statistics:

COLLECT SUMMARY STATISTICS ON <TABLE>;

Table level statistics are the information about the number of rows per table, the average data block sizes, and the average row sizes.

The number of rows per table is a very important measure for the Teradata Optimizer to estimate the cost of a full table scan. So, how was this information made available in previous releases of Teradata?

In previous releases of Teradata, this was achieved by collecting column statistics on the dummy “PARTITON” and the Primary Index columns.

Summary statistics which are defined on a table, will be refreshed each time the statistics on any column of the table are refreshed. In order to avoid the waste of resources, the best practice is to refresh all column statistics at once instead of doing it column by column:

The preferable solution is this one:

COLLECT STATISTICS ON <TABLE>; — all column statistics are refreshed, summary statistics are only refreshed once.

Although less intense on resources than traditional column/index statistics, this should still be avoided:

COLLECT STATISTICS ON <TABLE> <COLUMN_1>; — first collection of summary statistics

COLLECT STATISTICS ON <TABLE> <COLUMN_2>; — needless 2nd collection of summary statistics

Conclusion: Summary statistics collection is a very useful feature to refresh the number of rows information per table quickly.

1. Dimitrios says

   May 7, 2014 at 11:07 pm

   Hello,

   This new feature has only advantages , except from one thing , the way that we have to change the procedure which collect the stats in order to use this feature(s) (plus the max value length / max internal and sample ).

   It is also recommended to big tables , where collecting stats consumes a lot of io/cpu , just
   Collect Summary Stats on DB.TB ; ,after every DML transaction on it , in order to have updated summary stats on the table , in this way we help the parser to make better extrapolations .

   Reply

2. Johne490 says

   August 20, 2014 at 6:23 am

   Very informative post.Really thank you! Awesome.

   Reply

3. Sreeraj says

   March 30, 2015 at 7:23 pm

   Thank you. very informative.
   Question: for TD 14 onwards, the best practice is to collect all column statistics at once instead of column by column. is this true for big tables as well? I was under the impression that we need to collect stats only for certain columns based on the PI / join and WHERE. could you please clarifiy?

   Reply

4. Roland Wenzlofsky says

   March 30, 2015 at 7:31 pm

   Hi Sreeraj. I think this is a misunderstanding.

   “All columns at once” means to collect statistics with the new syntax which allows to collect the stats on different columns of the same table at once.

   Before Teradata 14 we had to collect stats column by column i.e. issue one statement per column. The only performance advantage previously possible was synchronized scanning on the spool (this was achieved by starting the collect statistics statements for one table at the same time).

   “All columns at once” does not mean to collect statistics on each and every column of a table…

   Roland