how optimize importation ?

["Pierre FERT" <pfe@calo.ath.cx>]

Hello,
 I seek to make an importation in openldap of more than 3.000.000 of entries
within the shortest times. 
Then well on I have some questions has to pose to you.
 The importation is carried out starting from a ldapadd and not a slapadd
because I do not have the data timestamps UIDS etc?.
How can I to make differently? I noted to make an indexing after an
importation LDIF is faster but, I have the impression that the objectclass
are indexed in any event has the importation even if `index objectClass eq'
is not specified in slapd.conf. 
What wants to thus say that the objectclass are still indexed by slapindex,
it is thus a waste of time, how can I to make differently?
 To optimize the importation I have: 
Slapd.conf :
Backend bdb
Loglevel = 0
put that 3 shemas necessary A the importation (core/cosine/private) 
schemacheck     off
dbnosync
dbnolocking 
# replogfile
disabeled the monitor and indexs DB
DB_CONFIG :
set_flags DB_TXN_NOSYNC (but it seems to me that dbnosync of slapd.conf is
equivalent) 
set_cachesize 0 268435456 1 (for 1Go of memory 512 was slower)
and I did not modify the repertory of the logs because I nothing gained
there, I found that odd besides

since I make db_archive - D to remove the whole of the logs, would not have
it there not a means of avoiding creating them? 
kind to make an importation with a backend which is faster to make a
slapcat. to go up a backend bdb and to import the LDIF directly in the DIB
without passing by the DSA?

Thank you for your councils.

and sorry for translater