Load balanced configuration

[markdcc]

I am trying to configure two JBoss/Linux servers to handle web requests while accessing the same storage filesystem and database.  The outcome is to mainly provide higher performance with a bit of redundancy on the front end.  The database server (MSSQL) and file system (SMB share) will be on separate servers.  We have a load balancer device which will direct incoming requests to either server, but maintain persistence between each client session and a particular server.

We have too much data to replic_ate the entire repository.

My questions are:

Will two Alfresco servers pointing to the same database and alf_data folders cause concurrency issues or data corruption?

Do I need to perform cache replic_ation between these servers?

Are there any instructions that can help me along the way (other than the High Availability Configuration wiki entry)?

Thank you for any help you can provide.

(I added the underscores to get past the forum spam filter)

[derek]

Hi,

Everything can be shared except the Lucene indexes, and obviously the caches, both of which need to replic8.  Clearly there is a need for documentation on the 1.4 clustering - our partners and clients have samples, but I've not had the time to translate them for the Wiki.

Regards

[markdcc]

Thanks Derek,

That is pretty much what I had come up with.  I never did get the cache replic8tion to work, but I have both boxes with there own copy of the lucene indexes, running recovery every minute, and cache turned off.  Do you see any future headaches I might run into with this configuration?

[derek]

Hi,

The cache will become a performance issue.  If you find that the machines serve their requests fast enough without the caches, then you're lucky.  They make a big difference.  Under load, you might see the database struggling unduly as a result.  I'd say that it's a workaround until you get the EHCache clustered.  You can copy the ehcache-default to <extension>/alfresco/extensions/ehcache-custom.xml.  It will be picked up as the default automatically.  Then follow the instructions on the EHCache site.  Once you've done it once, it's done and will be worth the effort.  I want to make that config a single switch, but it'll require some low-level hooking into EHCache, so I've not had time, yet.

I'd also say that you should make the index recovery task fire more frequently.  5s or 10s at the most.  Just as long as it is on remoteOnly, it will not affect your system noticeably.

[markdcc]

Good tips!

I was struggling to get TreeCache working for the better part of a week, before giving up, and I have noticed some serious slowdowns.  It sounds like EH clustering will be easier, I will look into it.

Thanks again!

[derek]

Do you have sticky sessions on the load-balancer?

We have thrown the TreeCache out.  Apart from anything else, it didn't fit nicely into the Tomcat world.

[markdcc]

Yep…. Sticky sessions.