CTDB

[ronnie sahlberg]

On 8/17/07, Andrew B. Lundgren <lundgren at byu.net> wrote:
> Andrew Bartlett wrote:
> > On Thu, 2007-08-16 at 21:54 -0600, Andrew B. Lundgren wrote:
> >
> >
> >> The clusters themselves will be replicated, so that will provide one
> >> level of redundancy.  Within a DFS, I don't know yet.  I haven't
> >> research the DFS enough yet.
> >>
> >> So, if I set up two DFS shares in SAMBA that are configured as replicas
> >> of the same data, I will need to be using CTDB on the two nodes with the
> >> replicated DFS data?
> >>
> >
> > Yes.  I suppose in a very simple setup each of these pairs would be
> > their own cluster.
> Then, if I want to arbitrarily create replicated DFS shares across 4
> nodes in a DFS cluster, not using a clustered file system, the four
> nodes should use CTDB?

No, you shouldnt use CTDB of you build a DFS cluster.
You should only use CTDB if you want a cluster where you share out the
same namespace from all nodes at the same time.

That will give you a very scalable cluster since you can distribute
access to the same filesystem across all the nodes simultaneously.

It will also provide great redundancy since if a node crashes, you do
not lose access to part of the namespace. the only thing that happens
when a node fails is that you may get a small reduction in aggregated
throughput since the load will be shared across a smaller number of
nodes.