On Fri, May 06, 2005 at 08:27:53AM -0700, David Mathog wrote: Other points well taken, and one of the nice things about having the source code to modify. :)
<snip>
> > Have you looked at the multicast stuff instead?
> > That scales as O(1), and doesn't suffer the multiple
> > single points of failure.
>
> True. On the other hand the daisychain method can address
> a fraction of the nodes without bombarding the others with a lot
> of unwanted broadcasts. My cluster is homogeneous but in a
> heterogeneous cluster one might imagine loading different images
> onto each particular type of node - and leaving the others running
> at the same time. Ditto for a phased upgrade, in case only
> some small fraction of the working nodes at a time can be taken
> out of service for upgrades. That isn't my environment, but I
> can imagine sites that have requirements like that.
Multicast != Broadcast. If the clients don't sign up for the multicast
channel they won't see any packets.
-Sean
--
__________________________________________________________________
Sean Dague Mid-Hudson Valley
sean at dague dot net Linux Users Group
http://dague.net http://mhvlug.org
There is no silver bullet. Plus, werewolves make better neighbors
than zombies, and they tend to keep the vampire population down.
__________________________________________________________________
pgp1t0FR0q91m.pgp
Description: PGP signature
