Fwd: Re: Questions about GENI for XIA project

[Niky Riga]

Hi Matt,

I am forwarding your email to geni-users since we will need someone from
Utah to help along in debugging this. What I notice here is that your
topology is already reserved and you are trying to update the topology
by adding a node. I don't know if that is important for the error you
are getting.

You said that before the controller was at CMU, I guess that means that
the controller can talk to the nodes over the control interface and it
does not need a layer two connection to your nodes. If that is the case
then maybe you can try reserving the controller with no dataplane links,
but make sure it is in the same node (pc545) as the rest so that there
is a chance that although the control IPs are NATed that they can still
talk to each other (Utah folks do you believe that something like that
would work)?

If you do need a dataplane connection between your nodes and the
controller, I would suggest that you leave this slice as is and try and
bring up a new slice with the controller included and see how that goes.

Good luck,
Niky

-------- Original Message --------
Subject: Re: Questions about GENI for XIA project
Date: Thu, 23 Jan 2014 11:36:12 -0500
From: Matt Mukerjee <muke...@cs.cmu.edu>
To: Niky Riga <nr...@bbn.com>
CC: David Naylor <dna...@cs.cmu.edu>

Thanks Niky,
I've attached the topo as a picture below. It's really messy as my screen
resolution is pretty low. Basically we're making a three tier topology. The
top tier has two nodes (VM and VM-0), the middle tier has two nodes (VM-1
and VM-2), and the bottom tier has 6 (VM-3 thorugh VM-8). Each tier can
connect to every node in the adjacent tier (i.e., VM-0 can connect to both
VM-1 and VM-2; VM-1 can connect the above tier [VM and VM-0] as well as the
below tier [VM-3 through VM-8]). Setting up a network like that works fine.

We had gotten an email from IT guys at emulab saying that our experiments
are causing unusual traffic between emulab <--> cmu (as our controller node
is at cmu), so we thought we could move our controller node into emulab as
well. In the picture below this is VM-9. Since it's a controller, we'd like
it to be able to talk to all nodes. When you start connecting the
controller to the rest of the nodes you start getting that red line as seen
in the picture. This happens both if you create individual LANs for each
connection from controller to node, or if all of these connections share a
LAN.

Any ideas?

[Matt Mukerjee]

Hi Niky,

Thanks for forwarding this along. That's a smart idea to bring up a node on the same machine without data plane links. That would probably work!

As I mentioned previously, our traffic between emulab and our controller at CMU was flagged as unusual, so I'm trying to track down why that is first. In the best case I hopefully won't need to move the controller at all.
Thanks!

-Matt