[Linux-cluster] Cluster vs Distributed?

[Lon Hohberger]

On Tue, 2006-10-24 at 14:53 -0500, isplist at logicore.net wrote:

> Of course, if a machine needs rebuilding, then it's a full rebuild, then copy 
> what I need, make it what it's supposed to be and stick it back into the 
> cluster. Each server shares it's data with the others.

You should be able to automate this step, at least, mostly.  If you are
using Red Hat Enterprise Linux or Fedora Core, you can automate most of
the installation and configuration using Kickstart.

> Now on the other hand, the clustering I think the Linux LVS (is that what it's 
> called?) would give me a single single image type cluster would it not? I 
> mean, each node would be a part of a single system, just keep adding nodes to 
> expand the system and performance.

LVS appears to web (or other) network clients to be a high-capacity
single system.  In fact, you probably *are* using LVS right now -- it's
what is likely doing the load balancing.  LVS "real servers" (in your
case, the GFS cluster) do not have to be a part of a cluster, and in
fact, do not even need to know one another exist (of course, this
changes when you introduce GFS :) ).

Contrast to OpenSSI, which gives you a single system image for processes
running on the cluster:

  http://openssi.org/

> Which is best then? It seems each does similar things with it's own pluses and 
> minuses. Is my use of the GFS clustering a good use or am I missing serious 
> benefits?

Not all applications designed to work on a single system are instantly
scalable just because GFS is the back-end store.  Any application which
does not do proper file locking could cause problems for other instances
of that application elsewhere in the cluster, for example.

It all depends on what you're trying to do.  Could you draw us a
picture/diagram?  :)

-- Lon