[Nsi-wg] Topology properties for NSI - a minimalistic approach

[John MacAuley]

Evangelos,

I think your simple graph described in #1 will not hold and we will need 
more than a single link between nodes.  If we are supporting #5 then we 
could need multiple links to represent different values for the physical 
connections, especially if we are doing route diversity.  In addition, 
there my be multiple links as a result of topology at different layers 
between the nodes (unless these are considered separate graphs).

John.

On 10-02-10 8:09 PM, Evangelos Chaniotakis wrote:
> Hi all.
>
> Here is my take on the topology issue:
> - We SHOULD describe the topology properties that we think we will 
> need for pathfinding.
> - We SHOULD NOT assign too many semantics to the topology elements.
> - We MUST NOT design a new topology schema.
>
> Now for the things that I think are absolutely required properties for 
> any topology:
>
> 1. The topology is a directed simple graph.
> Note: Directed graph means each edge is an arrow with a start vertex 
> and an end vertex. Simple graph means there are no parallel edges i.e. 
> going the same direction between the same two vertices.
> (I am on the fence for the "simple" requirement; do we need to make 
> the topology a multigraph for some reason?)
>
> 2. We are allowed to separate the topology into mutually exclusive 
> subgraphs. I will use Jerry's term "Network Domain" for these subgraphs.
> Note: A subgraph is an arbitrary set of vertices and edges. It may 
> consist of only edges, only vertices, or a mix thereof. A Network 
> Domain is not necessarily a connected graph. (i.e. there is not 
> necessarily a path between any two of its vertices).
>
> 3. The union of all Network Domains MUST equal (cover) the entire 
> topology.
> Rephrased: a given vertex or edge MUST belong to exactly one Network 
> Domain.
>
> 4. A vertex belonging to a Network Domain MAY be connected to a vertex 
> belonging to another Network Domain. In that case, that vertex for NSI 
> purposes will be classified as a "Service Termination Point".
>
> 5. Edges and vertices MAY be annotated with attributes and metrics: 
> i.e. framing, capacity, cost, etc.
> Note: We MUST NOT specify those - this is a job for NML. But we SHOULD 
> specify what we think we will need from these attributes.
>
> 6. Each vertex and edge MUST BE addressable through a globally unique 
> identifier.
>
>
> Note that in the above there are very few hard requirements for the 
> topology graph - it can be as rich or as simple as necessary.
>
> I have intentionally avoided assigned any names (such as "node", 
> "port", "link", "point" etc) to the edges and vertices of the topology 
> graph.
>
> Finally, I think that the above requirements are enough to satisfy the 
> needs of pathfinding. Can anyone think of any further requirements? Or 
> is there a way we relax these even more somehow?
>
>
>
> Here are a couple of examples of graphs (attached also in .graffle). I 
> will work on more complex ones if needed but really, they will be 
> pretty much equivalent to the graphs on Jerry's slides.
>
>
>
>
>
>
>
> _______________________________________________
> nsi-wg mailing list
> nsi-wg at ogf.org
> http://www.ogf.org/mailman/listinfo/nsi-wg
>    

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.ogf.org/pipermail/nsi-wg/attachments/20100216/3caddacd/attachment.html