How to use index when use analytic functions?

Question

Oracle 11gR2, I have a an indexed snapshot log table (sequence$$ and PK indexed), and I use a query to access it...
query looks like:
SELECT "M_ROW$$", "SNAPTIME$$", "DMLTYPE$$", "OLD_NEW$$", "CHANGE_VECTOR$$", "SEQUENCE$$", "ID", "SITEID", "WEBPAGEID", "WEBFOLDERID"
FROM mlog$_table1
WHERE dmltype$$ = 'D'
AND sequence$$ IN
(SELECT DISTINCT
LAST_VALUE (sequence$$)
OVER (PARTITION BY siteid ORDER BY siteid)
FROM mlog$_table1);
how could I access mlog$_table1
explain plan looks like:
Plan
SELECT STATEMENT ALL_ROWSCost: 53 Bytes: 605.864 Cardinality: 1.736                          
     6 HASH JOIN Cost: 53 Bytes: 605.864 Cardinality: 1.736                     
          4 VIEW VIEW SYS.VW_NSO_1 Cost: 28 Bytes: 22.555 Cardinality: 1.735                
               3 HASH UNIQUE Cost: 28 Bytes: 45.110 Cardinality: 1.735           
                    2 WINDOW SORT Cost: 28 Bytes: 45.110 Cardinality: 1.735      
                         1 TABLE ACCESS FULL TABLE BONGO.MLOG$_PDSI1 Cost: 25 Bytes: 45.110 Cardinality: 1.735
          5 TABLE ACCESS FULL TABLE BONGO.MLOG$_PDSI1 Cost: 25 Bytes: 582.960 Cardinality: 1.735

Answer 1

ok, I think I feel like a newbie :)
explain plan for your query is:
Plan
SELECT STATEMENT ALL_ROWSCost: 22,1 Bytes: 416,678 Cardinality: 2,654           
     2 SORT GROUP BY Cost: 22,1 Bytes: 416,678 Cardinality: 2,654      
          1 TABLE ACCESS FULL TABLE IBIS.MLOG$_PARTIJA Cost: 22,098 Bytes: 416,678 Cardinality: 2,654
but it runs for about 3-4 sec..
1st query:
create or replace view x as
select * from mlog$_partija v
where not (DMLTYPE$$='D' and update_type=2)
and sequence$$=(select max(sequence$$) from mlog$_partija t
where t.par_sifpar=v.par_sifpar);
runs about 17 sec...
wich is expected since there's a max in subselect...
but, it's explain plan is:
Plan
SELECT STATEMENT ALL_ROWSCost: 108,957 Bytes: 1,84 Cardinality: 1                     
     5 HASH JOIN Cost: 108,957 Bytes: 1,84 Cardinality: 1                
          3 VIEW VIEW SYS.VW_SQ_1 Cost: 22,564 Bytes: 68.634.228 Cardinality: 2.639.778           
               2 HASH GROUP BY Cost: 22,564 Bytes: 68.634.228 Cardinality: 2.639.778      
                    1 TABLE ACCESS FULL TABLE IBIS.MLOG$_PARTIJA Cost: 22,05 Bytes: 68.634.228 Cardinality: 2.639.778
          4 TABLE ACCESS FULL TABLE IBIS.MLOG$_PARTIJA Cost: 22,145 Bytes: 4.788.557.292 Cardinality: 2.639.778           
but, I made a third solution wich is something like:
create table mlog_table1_last as
select siteid, max(sequence$$) max_seq from mlog$_table1 group by siteid
I run a query on that temp table:
select * from mlog$_table1 v
where not (DMLTYPE$$='D' and update_type=2)
and sequence$$=(select max_seq from mlog_table1_last t
where t.siteid=v.siteid);
it's explain plan is bad:
Plan
SELECT STATEMENT ALL_ROWSCost: 2.588.886 Bytes: 1,814 Cardinality: 1           
     3 FILTER      
          1 TABLE ACCESS FULL TABLE IBIS.MLOG$_PARTIJA Cost: 22,148 Bytes: 4.788.557.292 Cardinality: 2.639.778
          2 INDEX RANGE SCAN INDEX IBIS.MLOG_PARIJA_LAST_NDX2 Cost: 1 Bytes: 111,02 Cardinality: 4,27
but it runs for 1-2 sec..
why?