Comments on: Row sizes 2

By: raova

raova — Tue, 02 Oct 2012 15:26:38 +0000

10.2.0.3 has this 255 column restriction on compression so we haven’t tried that option.

By: Bernard

Bernard — Tue, 02 Oct 2012 14:50:20 +0000

Table compression will reduce such a lists of null columns to about one byte for all of them.

By: Timur Akhmadeev

Timur Akhmadeev — Tue, 02 Oct 2012 12:51:52 +0000

The “optimizer lady” said auto_sample_size is as accurate as 100%

… in Oracle version 11g or higher.

By: guest

guest — Mon, 01 Oct 2012 17:28:21 +0000

The “optimizer lady” said auto_sample_size is as accurate as 100%. Regardless, in this particular case, 95%+ accuracy is good enough for most people.

By: raova

raova — Mon, 01 Oct 2012 15:10:17 +0000

May be true, provided the stats gather was done with 100% sample size to get an accurate value for num_nulls.

By: guest

guest — Mon, 01 Oct 2012 15:01:52 +0000

An alternative approach to previous comment is if you were to run following query and create new order of columns, that would be just as good – provided you have gathered Statistics on the table. Then again, who doesn’t gather stats on their tables now a days? :)

  SELECT table_name, column_name, column_id, num_nulls, num_distinct
    FROM all_tab_columns
   WHERE owner = :owner
     AND table_name = :table_name
ORDER BY num_nulls, column_id

By: raova

raova — Mon, 01 Oct 2012 10:58:07 +0000

Very nice article. We have done a similar analysis on our datawarehouse and carried out a column re-order excercise. we have saved 1.2TB of space (Tables + Indexes) from one of the data marts. We have arrived at new column order using different stratgy where we have got the number of nulls in each column and then did a CTAS of the original table with the new column order. Here is the cut-down version of the steps we have performed.

select  /*+ parallel(x,16) */
count(decode(C1,null,1,null)) C1,
count(decode(C2,null,1,null)) C2,
count(decode(C3,null,1,null)) C3,
count(decode(C4,null,1,null)) C4,
count(decode(C5,null,1,null)) C5,
count(decode(C6,null,1,null)) C6,
count(decode(C7,null,1,null)) C7,
count(decode(C8,null,1,null)) C8,
count(decode(C9,null,1,null)) C9
from
T1 x
/

We have ordered the resulting counts in ascending order to arrive at the new column order (we have used excel to transpose the output and plot some graphs etc).

Column Name	Number of Nulls
C1		0
C8		0
C7		4763
C4		4763
C9		5230377
C6		11645479
C3		88483772
C5		252383865
C2		2028039667

Then we have CTAs'ed the table with new column order. The tables are huge and one of the table with 978 columns has gone down by 42% in size. Ananda