[Nsi-wg] Example routing policies in NSI

[Henrik Thostrup Jensen]

Hi

On Thu, 5 Feb 2015, John MacAuley wrote:

>       Link preferral. Many networks are connected through multiple links, but can also send traffic
>       over backbonetransit networks. And there can be more of these. Being able to priotize between
>       these is pretty important.
> 
> 
> This sounds like a stander link "cost" type of issue where you want some links to have a lower cost, and
> hence, get preferred over higher cost links.
> 
> A -- B -- C
> |    |    |
> \--- D ---/
> 
> In this example lets say D is the transit network and the preferred path from A to C.  If we used OSPF
> then it is a 50-50 split on which path gets selected since both routes are of equal distance (link cost
> == 1).  If we increase the link cost of the non-transit links, say A-B to a cost of 2, then the A-D-C
> path will be selected until it is no longer a viable path.
> 
> Obviously, this type of cost based mechanism can be tricky in that now we have  path A-D-B the same cost
> as path A-B, but using some of our other policies can stop us from routing to B via D.
> 
> Have you seen this addressed a different way?

In BGP this is done with MEDs. Basically you can set one on a prefix, to 
indicate how close the prefix is to a link. The other end of the link can 
then apply its own policies to those MEDs. AFAIK this can get pretty 
complicated, and some times things jus have to be done by hand and 
"carefully".

>       There are no cases with selective transit (well, only trivial ones). For instance with
>       something likes this:
>
>       A - B - C - D
>            \ E
>
>       A can reach C & E, but not D.
> 
> 
> "There are no cases" - do you mean this example does not need to be supported?

It definitely needs to be supported. I was referring to your slide set 
where there are no cases like this.

> In this case is B a transit with no termination option for A?

B provides transit to C+E, but the link AUP of the A-B link doesn't allow 
D (I'd like not to start pointing fingers here, but it is a real case).

> Whom is enforcing this policy?

A filters on B's reachability. One of the reasons cases like this are 
problematic is that it is possible to end up with asymmetric routing, if 
things are not done carefully. To prevent this, C could apply some filters 
on what gets announced to D.

> Is this a policy on source network A or is it policed on B/C?  The 
> reason I ask is if 
>
>       Furthermore we are currently discussing this case with another network/community:
>
>       A - B - C - D
>
>       A should be able to reach B+D
>       B should be able reach all.
>       C should be able to reach B+C
>       D should be able to reach A+B+C.
>       (at least it is symmetric)
>
>       That is, A and C are not exchanging traffic, but C provides transit to D, and D has transit
>       to A through B and C. Yay!
> 
> Are all networks aware of these policies

I cannot get this question to make sense. The answer is no, but it is 
important to realize that a network may not look identical from other 
networks.

> or is it selectively enforced by a subset?

Yeah.


     Best regards, Henrik

  Henrik Thostrup Jensen <htj at nordu.net>
  Software Developer, NORDUnet