Mikemail.Txt

File contents

Date: Mon, 14 Mar 2011 09:05:30 -0600
To: emulab-admins@googlegroups.com

I think there is a basic misunderstanding here, due to there being so many
networks involved.

 * All nodes, including boss and ops, need to be on a "control network".
   In the simplest form, this is a single VLAN (and single IP subnet) on your
   control network switch.  This is probably what you would use.  In the
   not-so-simple case, the control network is broken up into three VLANs
   (and IP subnets): the "node control net" (nodes), "public control net"
   (ops), and "private control net" (boss) connected via a router/firewall.
   This is what we do at Utah.  So when you see reference to the node control
   net it does NOT mean the experimental network, which maybe is where the
   confusion is.

 * All nodes, excluding boss and ops, are on the "experimental network".
   This is the set of switches that are dynamically controlled to interconnect
   the nodes.

 * Boss also needs a distinct interface connected to an experimental switch
   for use as the "hardware control network".  This is a separate VLAN 
   (and IP subnet) that contains boss, the experiment switches themselves,
   and any IP-controlled power controllers.  This is how boss talks to the
   switches themselves to configure VLANs, etc.

So in your case, you should have boss, ops, and one interface from the five
machines connected to "c2960-3".  That interface on boss is where dhcpd, tftpd,
etc. should be listening. This interface on ops is where nfsd, etc should be
listening.  This interface on the nodes should be the one on which PXE is
enabled and from which they will boot.

The other interfaces on the five machines should be connected to the other
four switches.

In summary, you need a dedicated, "full-time" control network interface for
all the nodes on the control switch.  You don't appear to have that right now.