Heavy load problems

["Angel L. Mateo" <amateo@um.es>]

Hello,

	We have a openldap (2.3.30, debian etch version 32bits) with 4 nodes 
(xen vm with one cpu core, Intel Xeon 3.20GHz and 2GB of RAM). It that 
farm we have two databases one with 121K entries (for authenticating 
users, with lots of connections) and other with 40K entries (with public 
information of our students, for example).

	Now we have migrated to two openldap (2.4.21, ubuntu lucid 64btis) 
farms. One with 4 nodes (xen vm with two cpu cores, Xeon E5450 3GHz, 2 
GB of RAM) for the authentication database, and other with 2 nodes (with 
the same resources than the other) for the other database.

	The problem is that in the authentication farm we are having lot of 
problems due to heavy load, although we are not completely migrate all 
clients to this new farm (we already have clients querying the old 
farm), but in the old farm we didn't have any problem (just some 
punctual problems long time ago).

	The only difference between the old and the new farm is that the old 
farm was replicating information with slurpd. The new one is a 
multimaster configuration. The configuration of the replica is:

mirrormode on
syncrepl  rid=11
          provider="ldap://<server1>.um.es"
          binddn="<replica user dn>"
          bindmethod=simple
          credentials=<creds>
          searchbase="<base of database>"
          type=refreshAndPersist
          interval=00:00:00:10
          retry="5 5 300 +"
          timeout=1

syncrepl  rid=12
          provider="ldap://<server2>.um.es"
          binddn="<replica user dn>"
          bindmethod=simple
          credentials=<creds>
          searchbase="<base of database>"
          type=refreshAndPersist
          interval=00:00:00:10
          retry="5 5 300 +"
          timeout=1

syncrepl  rid=13
          provider="ldap://<server3>.um.es"
          binddn="<replica user dn>"
          bindmethod=simple
          credentials=<creds>
          searchbase="<base of database>"
          type=refreshAndPersist
          interval=00:00:00:10
          retry="5 5 300 +"
          timeout=1

syncrepl  rid=14
          provider="ldap://<server4>.um.es"
          binddn="<replica user dn>"
          bindmethod=simple
          credentials=<creds>
          searchbase="<base of database>"
          type=refreshAndPersist
          interval=00:00:00:10
          retry="5 5 300 +"
          timeout=1

	This configuration is in the 4 farm nodes. We have checked indexes and 
both farms have the same.

	Another difference is that authentication database entries have some 
attributes obtained from the other database via dynlist overlay. In the 
old farm, because the two databases are in the same server we had no 
problem, but in the new one we have to configure this database as an 
ldap proxy to the other farm, with cache overlay used in this.

	Heavy loads appears randomly in all nodes (not always with all nodes at 
the same time). When this happens, slapd is using 100% (well... top 
reports 200% because we have two cpus), there is no swap, no iowait....

	Any idea of what could it be the problem? Is this multimaster 
configuration so heavy load?

--
Angel L. Mateo Martínez
Sección de Telemática
Área de Tecnologías de la Información       _o)
y las Comunicaciones Aplicadas (ATICA)      / \\
http://www.um.es/atica                    _(___V
Tfo: 868887590
Fax: 868888337