...

Motivation

As more and more institutions adopt DSpace as a repository platform, it will become increasingly important for DSpace to be able to scale well. Exactly what it means for DSpace to "scale" is up for debate, and should be eventually defined here: ScalabilityMetrics.

...

• No automatic failover or node promotion
• Trigger-based update propogation means that (eg) schema changes cannot be automatically propogated across nodes, and it is unable to replicate large objects
• It is unable to detect node failure
• No support for a multi-master replication topology (ie: there will always be a single point of failure)
• No support for connection brokering
  All of these things combined mean that I will not be looking into using Slony-I. If anyone has any positive experiences with using it, please update this section.

Sequoia 

Sequoia  

Image Added!Image:Sequoia-raid10.png!
Sequoia (formerly C-JDBC) is a drop-in replacement for JDBC. No code changes are required for DSpace to use Sequoia in place of the current Postgres or Oracle JDBC drivers. It is released under an Apache license.
Sequoia claims to support lots of really useful features, such as RAIDb (think RAID for databases). A real "killer" feature (that, incidentally, could prove useful even without any clustering) is the support for a heterogeneous db environment by performing on-the-fly query translation. This means we could (in theory) drop support for anything other than postgres, and stick Sequoia in between DSpace and the database, and let it do the query translation for us. Naturally, this will introduce a performance penalty, but I'm not sure how great that will be. I suspect it would be comparable to using something like Hibernate, except that we wouldn't need to change any code to support Sequoia.

...