drop table t;

create table t as
select
case when mod(rownum, 4) = 1 then 1 else 3 end as skew
, cast(‘x’ as char(500)) as x
from dual
connect by rownum >>>>>
– Html conversion does not seem to reproduce correctly the above query. Read T is populated with 100000 rows

exec dbms_stats.gather_table_stats(null, ‘T’, estimate_percent => 10 -, method_opt => ‘for columns size skewonly skew’)

col column_name format a8
col low_value format 99999999
col high_value format 99999999
col endpoint_actual_value format a10

——————————————————————————
select sample_size, blocks, num_rows, last_analyzed
from user_tab_statistics
where table_name = ‘T’
;

SAMPLE_SIZE BLOCKS NUM_ROWS LAST_ANAL
———– ———- ———- ———
10344 3634 103440 20-AUG-08

select column_name, sample_size, num_distinct, num_nulls, num_buckets
, histogram
, utl_raw.CAST_TO_NUMBER(low_value) as low_value
, utl_raw.CAST_TO_NUMBER(high_value) as high_value
from user_tab_columns
where table_name = ‘T’
and column_name = ‘SKEW’
;

COLUMN_N SAMPLE_SIZE NUM_DISTINCT NUM_NULLS NUM_BUCKETS HISTOGRAM LOW_VALUE HIGH_VALUE
——– ———– ———— ———- ———– ————— ——— ———-
SKEW 10344 2 0 2 FREQUENCY 1 3

select column_name, endpoint_number, endpoint_value
from user_tab_histograms
where table_name = ‘T’
and column_name = ‘SKEW’
;

COLUMN_N ENDPOINT_NUMBER ENDPOINT_VALUE
——– ————— ————–
SKEW 2562 1
SKEW 10344 3

– In range cardinality on skew distribution
explain plan for
select * from t where skew = 3;

select * from table(dbms_xplan.display(null, null, ‘BASIC ROWS’));

PLAN_TABLE_OUTPUT
———————————————————————————————
Plan hash value: 1601196873

——————————————
| Id | Operation | Name | Rows |
——————————————
| 0 | SELECT STATEMENT | | 77815 |
| 1 | TABLE ACCESS FULL| T | 77815 |
——————————————

– Out of range cardinality on skew distribution
explain plan for
select * from t where skew = 2;

select * from table(dbms_xplan.display(null, null, ‘BASIC ROWS’));

PLAN_TABLE_OUTPUT
———————————————————————————————
Plan hash value: 1601196873

——————————————
| Id | Operation | Name | Rows |
——————————————
| 0 | SELECT STATEMENT | | 12810 |
| 1 | TABLE ACCESS FULL| T | 12810 |
——————————————

I haven’t upgraded any of my 10g databases to 10.2.0.4 yet, so I can’t test any ideas; but one possibility is that there are two conflicting strategies here – one is the “not present” strategy for frequency histograms, the other is the normal “out of bounds” strategy for dealing with values outside the low/high values.

The cardinality of one for the input of 4 could be because the legal range is 1 – 2, with a range of 1; and the value 4 is outside the legal range by twice the size of the range.

Try repeating the experiment with legal values 1 and 10, and query for values inside the range (2,5,9 say), then a couple of value outside the range (12, 15, 21, 40 say) to see if the “half the min” rule is obeyed when you are inside the range, and some form of ‘tapering’ effect appears as you go outside the range.

But it seems to fail: look at below testcase there is a frequency histogram for column SKEW and when an out of range value for SKEW is used in the equality predicate (the last query), Oracle estimates cardinality of 1; due to included in 10.2.0.4 5483301 patch I would expect a cardinality of about (2491 * 10) / 2 = 12500 !

So what is wrong ?

Thank you for your comment.


Connected to:
Oracle Database 10g Enterprise Edition Release 10.2.0.4.0 - 64bit Production
With the Partitioning, OLAP, Data Mining and Real Application Testing options

------------------------------------------
| Id  | Operation         | Name | Rows  |
------------------------------------------
|   0 | SELECT STATEMENT  |      |     1 |
|   1 |  TABLE ACCESS FULL| T    |     1 |
------------------------------------------
rollback;