I am working with a clustered environment where a few instances of an application will be sharing the same index. In order to save space on the machines where the applications reside, I want to be able to store the index files on a shared mounted drive. Then I want to point each application to that shared drive for index access. Are there any issues with this approach (i.e. - delayed response time, synchronization issues, etc..)? The data set will be in the hundred thousands.

Technically it should be possible. Of course access over the network is always a little slower than a local disk. Synchronization wise the access to the Lucene directory will be protected by Lucene's own locking strategy. In the latest Hibernate Search release you can even configure the Lucene locking scheme. Depending on the amount of updates index updated might become a bottle neck so. An alternative could be a clustered Search setup using JMS or JGroups.

Another question is how you mount the shared drive. Are you for example using NFS? I vaguely remember reading somewhere that this could cause problems. You should check the Lucene forums for that.

--Hardy

Thanks for your response. I think I have also read in forumsn to use SAN over NFS. We have not decided how to mount the drive yet..

Hi,
if you have each instance writing updates to the shared index you could have problems with locking; especially NFS can't guarantee correct working of the file locks used by Lucene.
You have two options:

use the JMS approach (or similar approach using JGroups backend since latest beta release), so that you have only one single node doing the writes and the others doing the reading. Application would scale fairly well because you offload the indexwriters from the main nodes, and actually the single writing node can be setup as having "exclusive write access", which makes it perform better.

An alternative is to use a different LockFactory; I've written about the different LockFactories available on Lucene's wiki: http://wiki.apache.org/lucene-java/AvailableLockFactories
We're currently working on an Infinispan based Directory; it's too early to use the whole Directory in production but you could use the Infinispan approach for Locking only, it's extremely safe and efficient.
Of course you're welcome to try the Directory too, it should work AFAIK and needs feedback. It will definitely make the operations and management easier than shared filesystems, and theoretically perform better.

You mention that you want to share indexes to "save space" ..?? How big are the indexes? And incidentally you can't set it up to use a local index in a clustered environment, or each node will end up having a different incomplete index.