I'm interested to hear more about what you mean by complexity?

I've heard of some large layer 2 networks. They are "simple" (aka "familiar") but also have some operationally "complex" aspects.�

For the sake of argument, what is more "complex" about directly programming a forwarding element to do exacrtly what you want? Why is that complexity?


On Wed, Jan 9, 2013 at 12:51 PM, Kris Price <nznog@punk.co.nz> wrote:
VPLS is not hard, it's pretty simple, assuming you mean RFC4761 or RFC4762. It's not necessarily cheap to get good kit for it though.

If you want a Layer 2 fabric that bilateral peering can take place on without involvement of the IX, you could build yourself a public L2 service using SDN.

You could also have your big scale out router running on the fabric, and still do bilateral peering over Layer 2 E-Line services that are set up, with the IX presenting an easy to use portal for configuring these. I go to the portal and log in and say I want an E-Line to my peer X, and X goes in also to say he accepts, and we're both given an C-VID to tag our traffic to each other with. Anything I send to my port on the IX with that CVID pops out his port with the CVID he's given (potentially these are the same at both ends but doesn't have to be that way). Then we get billed a little bit extra at the end of each month. This is also achievable with SDN. Again it's just analogues to existing services, but set up and managed in a different, easier to automate way.

What's certainly true though is that the IX can make use of lots of those commodity 64x10GE switches, each costing sub 10,000 USD, to offer their customers pretty much infinite bandwidth (certainly in NZs case, one of these would happily carry all NZ's traffic right?), so why add complexity if it isn't needed and bandwidth at an IX can be made so cheap? Maybe one reason to go down that path is if you want the IX not to be just an *I* exchange but a general Ethernet exchange, which certainly might have some merits.



On 1/8/2013 2:20 PM, Sam Russell wrote:
The hard part of VPLS/MPLS/multiple VRFs over MPLS is getting the
routers to all agree on their view of the world, and having one-way LSPs
can complicate things further. If you have one controller that pushes
out a single unified view of the world to every device, you make it a
lot simpler, and therefore a lot harder to break things.

The current implementation (as I understand) pushes flows out
pre-emptively - the software RouteFlow syncs the RIB of a running linux
VM to the flow table on the switch

_______________________________________________
NZNOG mailing list
NZNOG@list.waikato.ac.nz
http://list.waikato.ac.nz/mailman/listinfo/nznog