[gomd-devel] automating clustermask in gomd

[Mirko Caserta]

Hi there,
I was talking with tantive on #openMosix a few minutes ago (yeah, at least from
the office I can connect to freenode.net ;) ) and we were thinking about how to
automate the clustermask thing for an heterogeneous cluster. This is a major
concern for Gentoo users as you know, because we have all apps recompiled from
scratch with architecture specific optimizations. At the same time, it might be
a problem for those who are using binary distroes as well. Think about an i586
MDK which has to migrate jobs in a cluster with i386 or i486 machines (yes,
there's a lot of such clusters out there ;) ).

How could we implement such a feature in gomd? I know gomd could be the right
place to do this because the autodiscovery daemon's job is supposed to only
provide a map of the cluster, not the features of each node.

At the same time, as I was saying in the oM-devel mailing list, I don't think
setting the clustermask is a simple thing of grepping the CPU flags in
/proc/cpuinfo. We need to understand what is needed in order to automate the
masks in a reliable way.

It is true that we can implement the mask thing in the monitoring application
(such as openMosixView, just to name a favourite one ;) ). But what if you
can't run a full-blown X11 app?

I mean, one of the major upsides of openMosix is that you don't need to do
anything special to make it work. If we could do the same to automate masking
nodes in an heterogeneous cluster it'd be simply great.

I have no idea where to start from btw. Maybe some of you might have some clues
about this.

Ciao, Mirko.