Setting up separate network question

[Ben Scott]

On 17 Mar 2008 13:33:59 -0400, Kevin D. Clark <kevin_d_clark at comcast.net> wrote:
>  I think that the easiest way to setup a cluster in the manner that you
>  describe would be to ignore ipmasq, let the nodes in your cluster get
>  their eth0 IP addresses via DHCP, and just statically assign some
>  private IP addresses to your node's eth1 addresses.

  I'm not following the above.  Where is the DHCP server for his
cluster coming from?  And what good does having private addresses on
his node's eth1 interface do?  And how do the cluster nodes reach the
Internet?  Are you thinking of having his node get multiple
Internet-facing addresses and doing proxy arp?

  For within the cluster: I find that DHCP is so much nicer than
manually configuring everything, especially when things change (like
they always do), or a computer fails and needs to be replaced.
dnsmasq can easily provide name resolution tied into DHCP to eliminate
the need for manually propigating host files (yuck!), and there are a
few different ways to let DHCP assign static addresses if you want
that.  Done properly, you can drop a new node into a network and have
it auto-install and auto-configure itself with a handful of
keystrokes.

  Granted, I've never worked on a compute cluster, but I don't know of
any reason why these tools shouldn't work there, too, assuming
commodity hardware (and it sounds like we are).

-- Ben