[DGD] DGD/MP & OpenMosix revisited?

[Ragnar Lonn]

On Mon, 9 Feb 2004, Felix A. Croes wrote:

> I was assuming high quality PC-level components.  I had no idea that
> they were as reliable as you say.  I take it that's mean time between
> failures for the cluster as a whole, not for each machine individually --
> that makes each component in a cluster machine several times as reliable
> as each component in a large machine, assuming that the latter has
> fewer components overall.  That's impressive.

Well that's obviously not true. A cluster with more components means more
hardware failures. But hardware failures is not a measure of reliability
of the system. The interesting things to measure are total cost (buy,
own/maintain) and level of service offered to the end user (uptime,
transaction time).

A cluster will always demand more of the software run on the system than a
single-server solution will. If the software is up to the job, the service
will run faster and more securely than it would have on a single server
system. But that is IF the software does everything right. If it doesn't,
then a multi-server system will not solve anything.

In my opinion, the question in this case is whether or not it's worth the
trouble writing the software, or if there are reasons to believe it can't
be done (well). Felix seems to think it is quite difficult but we haven't
heard whether or not it's more difficult than writing DGD/MP?

Regards,

  /Ragnar

_________________________________________________________________
List config page:  http://list.imaginary.com/mailman/listinfo/dgd