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

Re: Abysmally poor write performance

On Sat, Jan 18, 2003 at 09:24:32PM -0500, Jason Tackaberry wrote:

> Using RedHat 8's OpenLDAP 2.0.25, I have populated an LDAP directory
> with 200 entries.  The total size of all gdbm files is about 1500kb. 
> Each user (entry) has 7 objectClasses: top, person,
> organizationalPerson, inetOrgPerson, posixAccount, shadowAccount,
> qmailUser.  slapd is configured to maintain these indices:
> index   objectClass,uid,uidNumber,gidNumber,memberUid   eq
> index   cn,mail,surname,givenname,mailAlternateAddress eq,subinitial
> The system is a 1GHz P3 with 256M of RAM and SCSI disks (it is a Dell
> server).  Updating a single entry with GQ can take up to 5 seconds to
> write.  I just now wrote a script to update gidNumber on the 200 entries
> using ldapmodify(1), and it took 5 minutes 9 seconds.

That sounds much too slow. I tested 2.0.7 a couple of years ago, and
found that it could add 1000 entries in 290s with four attributes
indexed (the test was on a 600MHz Athlon PC with a single 5400rpm
disk, and I did not put any effort into optimisation).

I suggest you run 'top' while doing the add, and look for anything
chewing up the CPU. You could also try 'iostat -x 5' and watch for
overloaded disks.

With performance as bad as you report, I suspect two things:

1)	You might have a badly-configured DNS which is taking a long
	time to resolve addresses into names.

2)	You might have the log level set very high on slapd, resulting
	in most of the CPU and disk performance being consumed by

|                 From Andrew Findlay, Skills 1st Ltd                 |
| Consultant in large-scale systems, networks, and directory services |
|     http://www.skills-1st.co.uk/                +44 1628 782565     |