Performance with large CMS repository / cache preloading

[frontline2]

I noticed that the portal takes quite long to start up and eats a lot of memory when you have large amounts of data stored in the CMS.

I tracked this down to the class "JBossCachePersistenceManager", which apparently preloads stuff on startup into its caches.
A couple of questions:

- does this class actually read all the data from the database and cache it in memory?
- is it safe to disable this preloading? An if it's safe what is the correct way to do it (now I just commented out the preloading in the code and the portal seems to be working fine).
- why is this preloading needed in the first place and can the amount of stuff cached be adjusted somehow, this solution really doesn't scale that well...

Here are some performance numbers from the preloading, the strings are the hql:s run. The number is the time in milliseconds and the rows are the rows read from the database.

21:48:16,812 INFO [STDOUT] Time for: from org.jboss.portal.cms.hibernate.state.WSPRefs - 172, rows: 0
21:48:24,062 INFO [STDOUT] Time for: from org.jboss.portal.cms.hibernate.state.VersionRefs - 7422, rows: 4695
21:48:24,515 INFO [STDOUT] Time for: from org.jboss.portal.cms.hibernate.state.WSPNode - 7890, rows: 6076
21:48:33,828 INFO [STDOUT] Time for: from org.jboss.portal.cms.hibernate.state.VersionNode - 17219, rows: 36274
21:48:47,296 INFO [STDOUT] Time for: from org.jboss.portal.cms.hibernate.state.WSPProp - 30734, rows: 110958
21:49:32,828 INFO [STDOUT] Time for: from org.jboss.portal.cms.hibernate.state.VersionProp - 76266, rows: 319954

[soshah]

There should be no side effects if pre-loading is turned off.

The pre-loading is done, to improve the performance of JackRabbit requests for resources thats are stored in the database. Without the pre-loading, a simple property read results in about 300 database queries.

This is due to an inherent design issue with JackRabbit since the data is not structured ideally for storing in a relational database.

From an end user standpoint, without the pre-loading you will notice that the built-in CMS Admin tool is significantly slow

Thanks

[pmn92]

Could you tell how to turn it off please?

thanks

[antoine_h]

I am also interested in how to turn it off.
or tune it.
I will look also if there is something about that in the documentation...
may be write a wiki about the cache management in the cms ?