[Date Prev][Date Next] [Chronological] [Thread] [Top]

Re: slapindex, 1 million entries, some numbers



--On Saturday, May 31, 2003 11:08 AM -0300 Andreas <andreas@conectiva.com.br> wrote:

Well, actually I don't have many numbers. It's more than 12 hours now and
slapindex still hasn't finished. I already had to clean up transaction
logs twice, otherwise they would be at 9Gb now.

I'm making the following indexes:
index   objectClass     eq
index   cn,sn,mail      eq,sub
index   uid,givenName   eq
index   uidNumber,gidNumber     eq

These indexes may be too much, I don't know really, I just thought about
some search filters that I would like to be able to apply.

This is the slapindex line in top right now:
 3056 root      12   0  376M 239M 13124 S     0,3 27,1 155:01 slapindex

Now, this is a different machine, it's a quad PIII 700MHz with 2M cache
for each CPU, 8Gb RAM and two SCSI disks. Just for comparison, this one
took about 18 minutes more to run that slapadd with 1 million entries:

real    48m7.345s
user    17m32.690s
sys     3m44.530s


My DB_CONFIG:
set_lg_dir      /storage/ldap
set_lg_max      104857600
set_lg_bsize    26214400
set_cachesize   0 1048576 0


Hello Andreas,

I filed a bug a while ago against slapindex because of how long it takes to run. Howard Chu did a bit of analysis on slapindex, and what it comes down to, is that you either need 3 HD's, or you need to have shared memory enabled and 2 HD's. Otherwise, it is quicker to simply dump your DB with slapcat and then reload it with slapadd (assuming you have 2 HD's).

--Quanah

--
Quanah Gibson-Mount
Senior Systems Administrator
ITSS/TSS/Computing Systems
Stanford University
GnuPG Public Key: http://www.stanford.edu/~quanah/pgp.html