Re: [Nml-wg] Topology Model and Pathfinding
Hello John, I'm replying to the NML list, because I strongly believe that this issue must be discussed there. I think I understand your point now, I've created a diagram (attached) that I think describes the situation. The top view shows the physical topology, there are two networks, connected by an inter-domain link, which is owned by a third party. At the bottom I've tried to capture the way that you would like to describe this situation. Is that correct? I've attached the graffle file too so that you can correct it if I'm wrong. Jeroen. On 10/02/2010 09:51, John Vollbrecht wrote:
Hi Jeroen --
I cc some others in case they have comments.
The issue is what I was discussing with you at JTs. I will try to describe it better over the next few days. I certainly agree (as I said when we discussed this) we should talk about it at OGF or before.
Let me try a bit to describe the issue now - comments will be helpful in getting a good description.
I think the basic thing is that the atomic element in NSI is the set of resources that is controlled by a NRM (network resource manager). Those resources include switches/nodes a links in your terminology. For sake of argument call such a set of resources a network.
I think network could be covered in NML by calling such a set of resources a group or a network group. For NSI (as I understand it) edge of a network is a port, where a port may be on either a link or node in NML terminology. This is where NML topology is missing a terrm - the concept of a port at the end of a link. If that were there we could have a simple mapping between NML and NSI concepts.
Networks interconnect by joining ports together at a point. A point might be considered to be where a male and female connector join - either physically or logically. This is essentially the ASON (ITU G.8080) model for interconnection of subnetworks. If NML has the concept networks connecting at ports then I think there would be good mapping between NML, NSI and ASON.
If we use this model, then the NSI topology is of a set of networks interconnected at points (or by connecting ports). By definition network can (potentially) make a connection between any two points - so from a graph point of view, a network is a vertex and a point or joined ports are edges.
Some conversation -- To help resolve this and help possible conversation about it, I offer my understanding of the NML issue with this. Please correct me if I have it wrong. From a topology point of view NML assumes both ends of a Link are the same and can be identified by the node to which they connect. Adding ports to the end of links increases the number of items that must be defined in a topology which makes definition of a topology longer and longer. Further, since topology is well defined as between node and link it is not clear why treating networks as supernodes with links between them doesn't work.
There may be some way to combine these that I haven't thought of or understood yet. The problem I see from a inter-network (NSI/ ASON) point of view is that links are resources that are part of one or another network, not something outside them. From a reservation and provisioning point of view each network must make a connection between its ports. Ports from two networks are connected in a topology so when connections across the networks are provisioned, the concatenated connection across both networks is provisioned.
Perhaps there is a way that topology and resources can be separated so that links don't have to have names and are implied. From a provisioning and monitoring view the control is certainly in a node, and so there may be a difference between what is the resource termination point and the control points.
The one problem that having implied names doesn't seem to be able to handle is when a link is a resource independent of nodes at either end. In this case the link is a network which must be included in the topology for resource reservation and scheduling even though it is not included in the control topology.
--- I am interested in comments and suggestions - of course. I think this is a very important concept to work through between groups.
John
On Feb 10, 2010, at 11:07 AM, Jeroen van der Ham wrote:
Hi John,
The NML set out to create a topology model that is sufficient to do pathfinding, at least on the topological level. I understand from the discussion with you that you feel that NML is currently not providing enough terminology to fulfill the needs of NSI pathfinding. I think it is important that this is discussed in the NML group. However I don't feel I have enough understanding of your problem to make a good case for it. Would you mind explaining your problem to the NML mailinglist?
Thanks, Jeroen.
Hi Jeroen -- thanks for including NML-- I think we should also include NSI, but I think for the short term the folks that are most interested are also in NML. I added to your pics and included below. Your bottom picture is correct (second in the revised doc). The top picture has a link outside a network which doesn't fit my model, but I think is the way you think about it. I add a third pic to show that a network might terminate at the end of a link. What do you think? John On Feb 10, 2010, at 2:10 PM, Jeroen van der Ham wrote:
Hello John,
I'm replying to the NML list, because I strongly believe that this issue must be discussed there.
I think I understand your point now, I've created a diagram (attached) that I think describes the situation.
The top view shows the physical topology, there are two networks, connected by an inter-domain link, which is owned by a third party.
At the bottom I've tried to capture the way that you would like to describe this situation. Is that correct? I've attached the graffle file too so that you can correct it if I'm wrong.
Jeroen.
On 10/02/2010 09:51, John Vollbrecht wrote:
Hi Jeroen --
I cc some others in case they have comments.
The issue is what I was discussing with you at JTs. I will try to describe it better over the next few days. I certainly agree (as I said when we discussed this) we should talk about it at OGF or before.
Let me try a bit to describe the issue now - comments will be helpful in getting a good description.
I think the basic thing is that the atomic element in NSI is the set of resources that is controlled by a NRM (network resource manager). Those resources include switches/nodes a links in your terminology. For sake of argument call such a set of resources a network.
I think network could be covered in NML by calling such a set of resources a group or a network group. For NSI (as I understand it) edge of a network is a port, where a port may be on either a link or node in NML terminology. This is where NML topology is missing a terrm - the concept of a port at the end of a link. If that were there we could have a simple mapping between NML and NSI concepts.
Networks interconnect by joining ports together at a point. A point might be considered to be where a male and female connector join - either physically or logically. This is essentially the ASON (ITU G.8080) model for interconnection of subnetworks. If NML has the concept networks connecting at ports then I think there would be good mapping between NML, NSI and ASON.
If we use this model, then the NSI topology is of a set of networks interconnected at points (or by connecting ports). By definition network can (potentially) make a connection between any two points - so from a graph point of view, a network is a vertex and a point or joined ports are edges.
Some conversation -- To help resolve this and help possible conversation about it, I offer my understanding of the NML issue with this. Please correct me if I have it wrong. From a topology point of view NML assumes both ends of a Link are the same and can be identified by the node to which they connect. Adding ports to the end of links increases the number of items that must be defined in a topology which makes definition of a topology longer and longer. Further, since topology is well defined as between node and link it is not clear why treating networks as supernodes with links between them doesn't work.
There may be some way to combine these that I haven't thought of or understood yet. The problem I see from a inter-network (NSI/ ASON) point of view is that links are resources that are part of one or another network, not something outside them. From a reservation and provisioning point of view each network must make a connection between its ports. Ports from two networks are connected in a topology so when connections across the networks are provisioned, the concatenated connection across both networks is provisioned.
Perhaps there is a way that topology and resources can be separated so that links don't have to have names and are implied. From a provisioning and monitoring view the control is certainly in a node, and so there may be a difference between what is the resource termination point and the control points.
The one problem that having implied names doesn't seem to be able to handle is when a link is a resource independent of nodes at either end. In this case the link is a network which must be included in the topology for resource reservation and scheduling even though it is not included in the control topology.
--- I am interested in comments and suggestions - of course. I think this is a very important concept to work through between groups.
John
On Feb 10, 2010, at 11:07 AM, Jeroen van der Ham wrote:
Hi John,
The NML set out to create a topology model that is sufficient to do pathfinding, at least on the topological level. I understand from the discussion with you that you feel that NML is currently not providing enough terminology to fulfill the needs of NSI pathfinding. I think it is important that this is discussed in the NML group. However I don't feel I have enough understanding of your problem to make a good case for it. Would you mind explaining your problem to the NML mailinglist?
Thanks, Jeroen.
<Port-point.graffle><Port-point.pdf>
On 10/02/2010 12:41, John Vollbrecht wrote:
I added to your pics and included below. Your bottom picture is correct (second in the revised doc). The top picture has a link outside a network which doesn't fit my model, but I think is the way you think about it. I add a third pic to show that a network might terminate at the end of a link.
The top picture is meant to represent the physical topology. And would indeed be very similar to the way that the NML describes it. So your third picture basically means that Owner B is equal to Owner C? We have not defined this yet, but I imagine that in NML you can relate an owner to any object, including a Port and a Link. How exactly does that not fit your use-case? Jeroen.
On Feb 10, 2010, at 4:50 PM, Jeroen van der Ham wrote:
On 10/02/2010 12:41, John Vollbrecht wrote:
I added to your pics and included below. Your bottom picture is correct (second in the revised doc). The top picture has a link outside a network which doesn't fit my model, but I think is the way you think about it. I add a third pic to show that a network might terminate at the end of a link.
The top picture is meant to represent the physical topology. And would indeed be very similar to the way that the NML describes it.
So your third picture basically means that Owner B is equal to Owner C?
We have not defined this yet, but I imagine that in NML you can relate an owner to any object, including a Port and a Link. How exactly does that not fit your use-case?
So take your second picture, with the link described as a network. I don't know how to describe the interface between the middle network and the end networks. There is no link between them, so how do I know they are connected? How do I describe this in a topology?
Jeroen.
On 10/02/2010 14:32, John Vollbrecht wrote:
So take your second picture, with the link described as a network. I don't know how to describe the interface between the middle network and the end networks. There is no link between them, so how do I know they are connected? How do I describe this in a topology?
This sounds like you are creating your own problem. You are purposefully not describing inter-domain links. Instead you are describing all that as networks which are connected at a single point, and then you have to create new terminology in order to describe what is going on there. NML is proposing to describe the networks, the inter-domain link with its ports, along with their respective owners. How does that not work? Jeroen.
participants (2)
-
Jeroen van der Ham -
John Vollbrecht