[Nml-wg] [NSI imp] Interesting deployment issue.

[Freek Dijkstra]

On 28-7-2015 20:43, John MacAuley wrote:

> We encountered an interesting issue on the AutomatedGOLE over the last
> week. [...]
> 
> iCAIR and Netherlight have set up a peering over a link that traverses
> WIX.  The bidirectional port on the iCAIR side is defined with a VLAN
> range of 1779-1798 while the Netherlight side is defined with a range
> of 4020-4039.  WIX performs mapping between the ranges without any
> involvement from NSI (WIX is not yet under NSI control). [...]
> 
> There are two solutions to this problem (assuming the VLANs must stay
> mismatched):
> 
> 1. Get WIX under NSI control and programatically control the switching.
>  In this case the topology changes from iCAIR - Netherlight to iCAIR -
> WIX - Netherlight and we get a new NML description that allows
> pathfinders to compute the WIX translation themselves.
> 
> 2. If we cannot get control of the mid switching point then the only
> alternative would be to modify the two ends (iCAIR and Netherlight) to
> do manual mappings for each VLAN.  This means more NML entries, but it
> would let the pathfinder figure out which VLANs are interconnected.
[...]
> 
> Each VLAN mapping would be modelled individually, with in and out
> unidirectional NML elements, as well as individual bidirectional
> elements.  A lot more work but is a way to get around the problem we are
> seeing. [..]

Hi John,

Very astute observation. You are completely correct, and this is
as-designed. Solution 2 works. It is not pretty, but that's just because
the reality aint pretty. You have to revert to low level fidgeting. I'm
at least pleased to hear that it works in the end. Of course, we can all
agree that if WIX implements NSI, that would be a lot prettier and easier!

Freek