[TriLUG] Linux Clustering (high availability) and file systems

[Jon Carnes]

On Tue, 2004-04-27 at 14:42, Tarus Balog wrote:
> Gang:
> 
> Okay, I want to make a particular file system highly available.
> 
> For example, suppose, just suppose, I had a directory called  
> "/var/opennms" that I wanted multiple machines to be able to write to.  
> So if I had two active machines writing to that file system, and one  
> died, I'd have a third machine that could come on-line and pick up  
> where the failed machine left off.
> 
> It has to be fast and reliable (so nothing like NFS). Has anyone worked  
> with SAN equipment where we could dual attach two or more machines over  
> SCSI or Fiber Channel?
> 
> How do Linux Clusters handle making data highly available.
> 
> Relevant links and RTFM suggestions welcome.
> 

I'm guessing that "Cheap" is also one of the criteria (or at least
"inexpensive").

I do this using NFS on a back-end private switched network that connects
multiple Intel boxes together. Sort of the equivalent of Blade server -
but using cheap Intel boxes. If you don't happen to like NFS as the
transport then use rsync and ssh - doesn't matter as it's all happening
on a back-end private network.

Will there be multiple writes to the same files/directories? If you are
mostly doing reads then there is no need for a SAN setup. Anything that
puts new data out to the directory structures in a timely fashion will
suffice.

Otherwise - A SAN is a nice idea.

Jon Carnes