Andrey Ivanov wrote:

Hi,

There may be several attributes of interest to you as far as the memory consumption is concerned (http://www.redhat.com/docs/manuals/dir-server/8.1/cli/Configuration_Command_...) : nsslapd-dbcachesize nsslapd-cachememsize for every backend (by default, your data is in cn=userRoot,cn=ldbm database,cn=plugins,cn=config) nsslapd-import-cachesize (used only during ldif import)

Start with nsslapd-cachememsize - make that as large as possible and minimize nsslapd-dbcachesize

You can adjust the corresponding values by monitoring the attributes like currententrycachesize or entrycachehitratio of cn=monitor,cn=userRoot,cn=ldbm database,cn=plugins,cn=config (http://www.redhat.com/docs/manuals/dir-server/8.1/cli/Configuration_Command_...)

You can also use the logconv.pl script to examine the access log to see what types of searches are being done and which are not indexed properly.

2009/6/26 Tim Hartmann <hartmann@fas.harvard.edu mailto:hartmann@fas.harvard.edu>

Hi!


I was spending some time today trying to make sure that I was
getting the most bang for my buck today an my replica's and I
notices two items of interest that I was wondering if anyone else
had input on!

Firstly, after creating a number of indexs, my performance seems
to be really good, the exception that I noticed was "finger" I
noticed that finger takes a couple of seconds to return the data
on RHDS whereas on OpenLDAP, it pops right now in real time! My
first though was that I was doing an un-indexed search, but I
can't for the life of me figure out what I might not be indexing
that I should be!

The second thing I noticed was that on my servers, which are
RHEL5, running 32bit OS's with the PAE Kernels, RHDS doesn't ever
actually address more then 3 gig of ram! I was looking through the
documentations, and it looks like by raising the "Maximum Cache
Size" I'll be able to allow RHDS to use more of the available
memory.. did I get that right?


Anyway, as always thanks in advance for all the help! This list
has been a tremendous resource for an application that keeps on
showing it's value in huge ways!


Best,

Tim

--
389 users mailing list
389-users@redhat.com <mailto:389-users@redhat.com>
https://www.redhat.com/mailman/listinfo/fedora-directory-users

---

-- 389 users mailing list 389-users@redhat.com https://www.redhat.com/mailman/listinfo/fedora-directory-users

Tim Hartmann wrote:

Rich Megginson wrote:

...

Andrey Ivanov wrote:

...

Hi,

There may be several attributes of interest to you as far as the memory consumption is concerned (http://www.redhat.com/docs/manuals/dir-server/8.1/cli/Configuration_Command_...) : nsslapd-dbcachesize nsslapd-cachememsize for every backend (by default, your data is in cn=userRoot,cn=ldbm database,cn=plugins,cn=config) nsslapd-import-cachesize (used only during ldif import)

Start with nsslapd-cachememsize - make that as large as possible and minimize nsslapd-dbcachesize

...

You can adjust the corresponding values by monitoring the attributes like currententrycachesize or entrycachehitratio of cn=monitor,cn=userRoot,cn=ldbm database,cn=plugins,cn=config (http://www.redhat.com/docs/manuals/dir-server/8.1/cli/Configuration_Command_...)

You can also use the logconv.pl script to examine the access log to see what types of searches are being done and which are not indexed properly.

So after playing with logconv a bit, it looked like finger was making this call in the logs...

[26/Jun/2009:10:59:36 -0400] conn=283289 op=-1 fd=80 closed - B1 [26/Jun/2009:10:59:36 -0400] conn=283289 op=2 RESULT err=11 tag=101 nentries=0 etime=1 notes=U [26/Jun/2009:10:59:35 -0400] conn=283289 op=2 SRCH base="ou=really,ou=long,o=name,dc=school,dc=edu" scope=2 filter="(objectClass=posixAccount)" attrs="uid userPassword uidNumber gidNumber cn homeDirectory loginShell gecos description objectClass" [26/Jun/2009:10:59:35 -0400] conn=283289 op=1 RESULT err=0 tag=101 nentries=1 etime=0 [26/Jun/2009:10:59:35 -0400] conn=283289 op=1 SRCH base="ou=systems,ou=services,o=hascs,dc=fas,dc=harvard,dc=edu" scope=2 filter="(&(objectClass=posixAccount)(uid=foo))" attrs="uid userPassword uidNumber gidNumber cn homeDirectory loginShell gecos description objectClass" [26/Jun/2009:10:59:35 -0400] conn=283289 op=0 RESULT err=0 tag=97 nentries=0 etime=0 dn="" [26/Jun/2009:10:59:35 -0400] conn=283289 op=0 BIND dn="" method=128 version=3 [26/Jun/2009:10:59:35 -0400] conn=283289 fd=80 slot=80 connection from 1.2.3.4 to 4.3.2.1

But even after indexing "uid userPassword uidNumber gidNumber cn homeDirectory loginShell gecos description objectClass" finger still responds slower then it does on comparison to the older openLDAP servers... where we don't do indexing on all of these attributes, AND still claims that I'm running an search that hasn't been indexed! I'm I missing something glaringly obvious?

You only need to index the attributes used for searching: (&(objectClass=posixAccount)(uid=foo)) You need an equality index on objectClass (which should already be there, it is one of the default indexes) and an equality index on uid (again, should already be there).

The problem is this: [26/Jun/2009:10:59:36 -0400] conn=283289 op=2 RESULT err=11 tag=101 nentries=0 etime=1 notes=U [26/Jun/2009:10:59:35 -0400] conn=283289 op=2 SRCH base="ou=really,ou=long,o=name,dc=school,dc=edu" scope=2 filter="(objectClass=posixAccount)" attrs="uid userPassword uidNumber gidNumber cn homeDirectory loginShell gecos description objectClass"

The notes=U and err=11 mean that either the lookthrough limit has been exceeded, or you need to increase your nsslapd-idlistscanlimit: http://www.redhat.com/docs/manuals/dir-server/8.1/cli/Configuration_Command_...

This is not a good search anyway - the client is basically asking for all entries that match objectClass=posixAccount which could be thousands or more - what does the client intend to do with all of those entries?

Thanks!

Tim

-- 389 users mailing list 389-users@redhat.com https://www.redhat.com/mailman/listinfo/fedora-directory-users

On Fri, Jun 26, 2009 at 02:14:06PM -0600, Rich Megginson wrote:

This is not a good search anyway - the client is basically asking for all entries that match objectClass=posixAccount which could be thousands or more - what does the client intend to do with all of those entries?

The finger command also tries to match your query against the first or last name of a user, as stored in the GECOS field of user entries. On Fedora and derived systems (including RHEL), this part of its function can be turned off with the -m switch.

HTH,

Nalin