[Nsi-wg] Topology virtualisation

[Gigi Karmous-Edwards]

Hi Freek,

I agree that the removal of non-GOLEs from the topology graph is an 
alternative to creating a constraint during path computation. I thought 
that the removal from the graph would be easier, similar to removal of 
failed links (due to availability etc ) during crank-back.  Having said 
that, in GIRRA, both technology and policy is taken into account but 
availability is not. This is because we do not collect availability 
information only relatively static  information about the topology, 
therefore reducing complexity and the number or required updates.

With reference to your other comments about policy: having an open 
policy GOLE makes path computation easier, since the fewer "policy-rich 
" domains one has in the computed path the better. IMHO, an ideal global 
path will consist of only the source and destination domains and the 
rest of the path will consist of policy-free GOLEs. Leaving the policy 
of the path to the two endpoint domains only. Does this make sense?

I do realize that this is different than traditional approach and I 
realize that today in our real world,  GOLEs are not all interconnected. 
The above statements are based on an "ideal" network.

Kind regards,
Gigi


Freek Dijkstra wrote:
> Gigi Karmous-Edwards wrote:
>
>   
>> [...] during path computation I remove all non-GOLE domains from the global topology graph [...]
>>     
>
> I am curious why you made this choice. Path computation can take any of these constraints into account:
> - topology constraints (is there a connection?)
> - technology constraints (is it compatible?)
> - usage constraints (is it available?)
> - policy constraints (may it be used?)
>
> Looking at the Internet, BGP is only concerned with the first and last constraint (topology and policy). Let's assume for a moment that in optical networks these are also the most important constraints, then GOLE is much less relevant for path finding (only topology is relevant as it has no policy on its own) than an other domain (topology and policy to take into account). Based on that, I would argue that it is better to leave out the GOLEs and only take other domains into account.
>
> Since you do the exact opposite (ignoring non-GOLEs), I presume you argue that other constraints are the ones that matter.
>
> Given the other mails, I presume that you only look at topology constraints, ignoring policy, technology and usage constraints.
>
> If so, I have two questions:
> 1. Is there a specific reason why you think that policy, technology and usage constraints can be ignored?
> 2. Why do you think that topology constraints can be ignored for non-GOLEs? Do you perhaps assume that non-GOLE topologies will never be connected to more than one or two GOLEs and never be connected to a non-GOLE directly? Is this true for larger backbone networks such as GEANT, NLR and Internet2 (I presume that Internet2 is connected to Pacific Wave, StarLight and MAN LAN).
>
> Regards,
> Freek
>
>   
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.ogf.org/pipermail/nsi-wg/attachments/20100625/3ede7cbc/attachment.html