AnsweredAssumed Answered

Performance degrades in clustered setup

Question asked by unknown-user on Feb 8, 2010
Latest reply on Oct 21, 2010 by martyphee

I have setup a cluster of two Jboss nodes and I am using apache web server as load balancer. I have depoyed Alfresco 3.1 on both the nodes of the cluster and they are using a common database.

I have observed that clustered setup of Alfresco is slower than standalone setup. I tried checking in around 20 files in a given folder and same operation takes twice the time on clustered setup as compared to standalone setup. Initially I thought  that since I am creating new content and one node of Alfresco must be notifying the other node about the content creation, so that is an additional overhead.
So I tried reading properties of files I have created. Even the read operation i.e. getting all the properties of a given file in cluster was taking around twice the time in cluster as compared to standalone Alfresco.

My all three machines i.e. two cluster nodes and the standalone alfresco machine are identical in terms of configuration i.e. Java Heap space, perm space and CPU speed.

I thought that bottleneck in speed could be due to load balancer. So I tried to access one node of cluster directly and even there performance was slower by around a factor of 2.

I tried the above steps for a single user so it couldn't be an issue of load on server.

Do I need to make some additional configuration in my clustered setup to improve performance?

Right now "dir.root" property on both the nodes point to a shared network drive and "dir.indexes.backup" and "dir.indexes" properties point to local directories on both the nodes. I have set "alfresco.jgroups.defaultProtocol" property as TCP and "index.recovery.mode" as AUTO.