[Discuss] Redundant array of inexpensive servers: clustering?

[bogstad at pobox.com (Bill Bogstad)]

On Tue, Apr 1, 2014 at 3:08 PM, Richard Pieri <richard.pieri at gmail.com> wrote:
> Derek Martin wrote:
>>
>> It really depends, but mostly it isn't.  One of my earliest gigs was
>> managing just such an environment, which mostly included a few custom
>> applications which were not designed to be clustered.  With the right
>> hardware, it's fairly trivial.  But the right hardware is expensive.
>
>
> Clustering non-HA apps can be trivial if those apps have few dependencies.
> For example, clustering Apache by itself is easy. You just need a storage
> resource, a network resource, and the startup and shutdown scripts.
>
> The complexity arises when service groups depend on other service groups. In
> a three-tier web app you have the web server, the application server and the
> database server. There are dependency chains that have to kept intact for
> the whole thing to work. Service migration needs to be planned so that you
> don't end up with all three service groups running on a single node. That's
> just one point in a simple three-node cluster. Consider expanding that to a
> half-dozen nodes with several dozen service groups with various dependencies
> to each other. Planning that out is a lot of work and there are few short
> cuts to be had.

I'm not sure that it is so much dependencies as (possibly hidden)
state.  In the case of an application delivered via
a web server, if all information required to respond to each HTTP
request is encoded (via URL, cookies, etc?); I don't think
it matters that much how many tiers you have.  Once the servers have
to start maintaining session state, it gets more complicated.
That's not to say that it doesn't take some coordination for each
level of multi-tier system to get to the next level, but it is much
easier
if you only have to do this on a per server basis rather then per session.

Bill Bogstad