Re: [help] Slice with OpenFlow switch in ExoGENI

[Niky Riga]

Hi Aaron,

I am moving the thread to geni-users since we will need the help of the
experts here.

On ExoGENI I would say that you should start with VMs and only if you
really need to move to bare metal hosts.

For OpenFlow, do you need a whole VLAN or if you get an IP subnet would
be ok? I know that all the EG nodes are connected to the hardware OF
switch through a statically provisioned shared VLAN, and you can get one
or more IP subnets.

There was a limitation on getting a new private VLAN on the OpenFlow
switch, which might have been fixed, but not sure. Someone from the EG
team can probably clarify some of these issues.

So more concretely if having on IP subnet on a shared VLAN using VMs is
an ok place to start, I can share some rspecs with you.

Cheers,
Niky

Gember, Aaron wrote:
> I would like to run some experiments with the IBM OpenFlow-enabled
> switches that are in the ExoGENI racks, but I am struggling to create a
> slice with the appropriate resources.
>
> I want a slice with bare metal node (VM might be okay) that has two NICs
> (or one if two isn't possible) connected to the switch. I then want to
> be able to install OpenFlow rules in the switch to forward between the
> two ports on the switch to which the bare metal node's NICs are
> connected. It doesn't matter which ExoGENI rack is used (although a
> lightly loaded one would be preferred), and I only need the slice in one
> ExoGENI rack (nothing across racks).
>
> I think I figured out how to allocate the bare metal node using omni,
> but I'm not sure how to allocate the OpenFlow part. Is there a tutorial
> somewhere that can help with this?
>
> Thanks,
> Aaron
>
>

[Niky Riga]

Hi Aaron,

Attached are two rspecs:
1. The request rspec for EG that will attach a VM to VLAN 1750 which
is the shared mesoscale VLAN
2. The request for OpenFlow that will request the flowspace for you.

If you want two interfaces on the same host to connect to the switch I
would imagine that you probably need them on different subnets because
if they are on the same subnet kernel might just do the forwarding for you.

The rspecs are for the renci rack but you can modify them and use them
on any EG rack.

Cheers,
Niky

Gember, Aaron wrote:

> Hi Niky & geni-users -
>
> We are trying to measure rule installation latency on a hardware
> OpenFlow switch, so scheduling in the virtual switch in the hypervisor
> may interfere with the measurements. Nonetheless, we can start with a VM
> and move to bare metal if out measurements are noisy.
>
> Also, I think a subnet should be fine to start with. If you can share,
> or point me, to some sample rspecs, that would be much appreciated.
>
> Thanks,
> Aaron