Index is not used (why and what should i do) table too big

Question

hi guys,
having a headache now. (on 10gR2 on linux system)
I used to have a large table and this is how i managed it. (i am on standard edition (thus i have no partitioning)
i would have a
1) cron job run daily to keep track of the lowest and highest ID of the table PER MONTH (inserted into a table)
2) below is the table
CDR     MONTH     01-AUG-11 00:00:00     94118236     97584656
CDR     MONTH     01-SEP-11 00:00:00     97581362     100573669
CDR     MONTH     01-OCT-11 00:00:00     100570865     103631203
CDR     MONTH     01-NOV-11 00:00:00     103629760     106497084
CDR     MONTH     01-DEC-11 00:00:00     106494085     107306335
so as you can see, for the month of dec for example, the lowest CDR is 10649408 and the highest is 107306335.
3) so everytime i write a sql on this table if i want to get data for NOV
select * from cdr where ID >= get_cdr_id('CDR','NOV','SMALLEST');
and it would be
select * from cdr where ID >= 103629760;
In the past, the index will be used (index range scan), but now the optimiser is doing a FTS and is taking too much time.
therefore i went to force the index
select /*+INDEX(C, CDR_PK)*/* from cdr C where cdr_id > 103629760 and cdr_id < 106497084;
and the COST -> 1678927 (CDR, INDEX_ROWID)
13158 (CDR_PK, RANGE SCAN)
-- the result return is fast and good.
without the index
select * from cdr C where cdr_id > 100570865 and cdr_id < 103631203
the COST -> 440236 (CDR,FULL)
-- but the result return is slow like anything
My question is
1) which cost should i look at ?
the one with index hint have much higher COST, but the results return are so FAST.
1678927 (with hint) compare with 440236 (FTS)
2) is using the index is the correct way, why isnt my optimiser using it ?! how do i make the optimiser used it ?
Regards,
Noob

Answer 1

Iordan Iotzov wrote:
About the clustering factor– I think there are two possibilities:
1) The “clustering factor” is generally correct, i.e. it represents the true value of how the data is “scattered” through the table. The clustering factor for the fragment of data you are interested in (where cdr_id > 100570865 and cdr_id < 103631203), however, is different (smaller).
2)Oracle did not compute the “clustering factor” correctly. It is not uncommon occurrence, particularly in 10g. This blog entry by Randolf Geist is quite informative - http://oracle-randolf.blogspot.com/2010/01/clusteringfactor-what-if-analysis.html. Comprehensive information on that topic is available in Jonathan Lewis’ book Cost Based Oracle Fundamentals (chapter 5).
You can make the index more attractive by setting lower “clustering factor” value using DBMS_STATS.SET_INDEX_STATS. If you choose to do that, you should review your whole stats gathering procedure, because the next gathering would most likely reset the “clustering factor” to its old (high) value.
Iordan Iotzov
http://iiotzov.wordpress.com/