[Gluster-devel] GLUSTER and Databases

[Gordan Bobic]

On Wed, 5 Aug 2009 14:32:29 +0500, Anton <anton.vazir at gmail.com> wrote:
> Very interesting, does anyone have any experience with 
> Postgres SQL Database (Or may be other DB) located in the 
> gluster? For example Postgres have DB as multiple files, and 
> so multi-node setup should give quite good performance. The 
> question is the reliability and other possible issues, as 
> maybe performance?

Assuming it works without corrupting the database (and it probably will
corrupt the database since AFAIK PostgreSQL has no support for external
file locking or awareness of database file sharing even between instances
on the same node, let alone different nodes), there are going to be
performance issues.

It is extremely difficult for a clustered/distributed file system to match
performance of a conventional local file system for most uses. What are
your requirements? Are you after more performance or redundancy? Typically,
for conventional loads, you will be better off with a failover setup, right
up to the point where your capacity is exceeded. Once you hit that point
you'll have to re-think your data use, probably on application level, but
you might get away with some kind of a database proxy for load balancing.
Once you exceed the limits of that approach you'll need to look into some
kind of a data partitioning approach. But none of that has anything to do
with the file system.

Either way, a file system will only help you if you are storage bound. How
big is your data? Do your indices fit in RAM?

Either way, it is unlikely that any database performance problems will be
solved by a cluster file system.

Gordan