Re: Load a database into memory

>> Maybe I should create an index on (num_editeur, num_site, num_emplacement)
>> ?
> 
> 
> Yes. Try to. This is the best choice.

Here it comes :

VACUUM VERBOSE ANALYZE accord_editeur;
NOTICE:  --Relation accord_editeur--
NOTICE:  Pages 7096: Changed 0, reaped 6790, Empty 0, New 0; Tup 447032: 
Vac 0, Keep/VTL 0/0, Crash 0, UnUsed 23768, MinLen 124, MaxLen 124; 
Re-using: Free/Avail. Space 721776/0; EndEmpty/Avail. Pages 0/0. CPU 
0.39s/1.67u sec.
NOTICE:  Index ae_tracking_idx: Pages 2300; Tuples 447032: Deleted 0. 
CPU 0.07s/0.90u sec.
VACUUM

EXPLAIN :
NOTICE:  QUERY PLAN:

Nested Loop  (cost=0.00..228.27 rows=1 width=56)
   ->  Index Scan using parametre_tracking_idx on parametre par 
(cost=0.00..2.02 rows=1 width=8)
   ->  Index Scan using ae_tracking_idx on accord_editeur ae 
(cost=0.00..225.50 rows=60 width=48)

EXPLAIN

So, no change for the cost...
Only for the number of pages...

So, if I can't more optimize my index, what else should I try ?
It uses too much CPU for me...