[Nsi-wg] Topology virtualisation

[Inder Monga]

This has been a good discussion. The issues brought up so far are (and  
my take based on responses as well as my own opinion) :

1. In Topology representation, is there a need to differentiate a GOLE  
from a Network/Domain?

     - the consensus seems to be a "No". It seems that topologically  
there is no difference between a GOLE and a Domain

2. I need to include a particular domain? How do I specify that to the  
path computing engine in NSI?

   - The inclusion of domains can be done via the same concept as an  
ERO. You can ask for a partial ERO that puts those particular path  
constraints on the path computation decision. You could specify that  
you want to use a particular GOLE using a partial ERO. NSI should  
support asking for paths with EROs specified (which version up for  
discussion)

3. There may be different policies associated with a particular  
network (domain, link, open-exchange, etc). How and where does that  
have bearing on the whole path computation process?

   - In the tree/chain model, each network domain being transitioned  
has the need to do their own intra-domain path computation. In order  
to do that, it has to apply the policies of the domain. The Service  
Plane architecture allows many different points in the reservation  
process to apply local policies. The first is when a user sends a  
service request to a NSA Provider Agent. The agent can apply policies  
based on egress/ingress ports requested through that domain and accept/ 
reject the connection. Similarly the PCE model allows for a series of  
decisions to take place based on constraints till a workable  
alternative is clear.

Distributing policies of links/connections between domains is  
inherently a very complex process and should be avoided. I would  
rather we focus on network capabilities instead.

Inder

On Jun 28, 2010, at 5:56 AM, Freek Dijkstra wrote:

> Gigi Karmous-Edwards asked:
>
>> So, how is this modeled? Victor suggests having the link between two
>> GOLEs as a single domain?
>
> This is a good approach IMHO. It allows one to add a policy to a link.
> And as Erik-Jan eloquently pointed out, a link can have a policy.
>
>> I think the key question that started this debate was: Are GOLEs
>> significantly different from Domains such that we need to model them
>> differently?
>
> Bearing the KISS principle in mind: If they can be modeled the same  
> way
> for path finding, then I would do so. That would make the architecture
> more simple, and thus more powerful.
>
> I have not yet seen a compelling reason to model them differently.  
> Such
> reason might still be there (I can imagine that it would speed up  
> things
> knowing beforehand that a GOLE has no internal bottlenecks or is  
> policy
> free.), however, I have not seem a simulation to prove that.
>
> Regards,
> Freek
>
>
> _______________________________________________
> nsi-wg mailing list
> nsi-wg at ogf.org
> http://www.ogf.org/mailman/listinfo/nsi-wg