Hello, I have been examing some of the topologies of the IDC network at Internet2 and others more closely and I have encountered some things that we may want to look at. One of the things that I would not know how to describe exactly in NML is the following example: A domain has a port on one of its devices. This port has an inter-domain connection. Through this port there are two possible connections to other domains, by means of separate VLAN labels. For example VLANs 1000-2000 go to one domain and 3000-4000 go to another. There are different ways of solving this at the moment. You do have to describe this somewhere as a single port, because of the limitations on the capacity of this port. I came up with three possible solutions, all with their own disadvantages: - Describe a single port with two links attached to it. This means breaks the assumption that a port can have at most one bi-directional link attached to it. - Describe a single port with two virtual ports each with its separate link. This breaks the assumption that a port is always connected by a Link (either a cross-connect or an external link), and we introduce a new concept of a virtual port. Are there other cases that would warrant this introduction? - Describe a single port that is connected to a "virtual" node where the connection is split into two. This makes the description more complex, and also seems to allow a direct connection between the two inter-domain connections without using the original port. Any other ideas? Jeroen.
Hi Jeroen, Jeroen van der Ham wrote:
One of the things that I would not know how to describe exactly in NML is the following example: A domain has a port on one of its devices. This port has an inter-domain connection. Through this port there are two possible connections to other domains, by means of separate VLAN labels. For example VLANs 1000-2000 go to one domain and 3000-4000 go to another.
Your example is not very clear. Where does the 'splitting' to to different domains occur: at the Node where this port resides, or at the other end of the connection? I mean an Ethernet port connects to one other Ethernet port, and if VLAN are switched to different domain this occurs within a device (in our case it would be handled by a SwitchingMatrix) In any case, can't this problem be solved by the schema you just posted yesterday? I think you could describe the above as follows. I would be curious to know what I misunderstood about the current schema... nodeA has port portA portA adaptation source ethPortA1000-2000 ethPortA1000-2000 adaptation sink portA portA adaptation source ethPortA3000-4000 ethPortA3000-4000 adaptation sink portA nodeB has port portB portA adaptation source ethPortB1000-2000 ethPortB1000-2000 adaptation sink portA portA adaptation source ethPortB3000-4000 ethPortB3000-4000 adaptation sink portA portA source linkAB portB sink linkAB portA sink linkBA portB source linkBA ethPortA1000-2000 source linkAB1000-2000 ethPortB1000-2000 sink linkAB1000-2000 ethPortB1000-2000 source linkBA1000-2000 ethPortA1000-2000 sink linkBA1000-2000 ethPortA3000-4000 source linkAB3000-4000 ethPortB3000-4000 sink linkAB3000-4000 ethPortB3000-4000 source linkBA3000-4000 ethPortA3000-4000 sink linkBA3000-4000 Here a little fuzzier as the schema is not so clear, so I 'make up' some terms (happy to hear how to handle this) nodeB hasService SwitchingMatrixB SwitchingMatrixB has port ethPortB1000-2000 SwitchingMatrixB has port ethPortB3000-4000 SwitchingMatrixB has port ethPortC1000-2000 SwitchingMatrixB has port ethPortD3000-4000 ethPortB1000-2000 switchTo ethPortC1000-2000 ethPortB3000-4000 switch ethPortD3000-4000 At this point you have splitted the vlans and proceed 'normally toward the to the two domains (via portC and portD in my example). So now, what do you think? Best regards, Paola ps = I am using ports with vLANs ranges and not one client port per VLAN....
Paola Grosso wrote:
Your example is not very clear. Where does the 'splitting' to to different domains occur: at the Node where this port resides, or at the other end of the connection? I mean an Ethernet port connects to one other Ethernet port, and if VLAN are switched to different domain this occurs within a device (in our case it would be handled by a SwitchingMatrix)
I'm sorry, it's a complex situation, and hard to describe. What I meant was that the VLAN tagging happens in the node itself, then both VLAN sets travel over the same Ethernet port and same cable. Outside of the domain this signal gets split, and both VLAN sets head of to their separate destination domains.
At this point you have splitted the vlans and proceed 'normally toward the to the two domains (via portC and portD in my example).
What you describe is what I meant with my third possibility, adding a virtual node. Yes, this node is there, but its not under the domains control, so no control on switching can occur there, other than what is configured now. The description you gave may suggest that it is also possible to directly switch between ports C and D. Your description also uses two different adaptations for the separate VLAN sets. AFAIK we haven't decided on how to describe this yet, either like this, or with a single adaptations with a possible label set attached to it. Also, this is the possible situation, how do we describe an actual configuration? Jeroen. PS. This problem, and many others like it is why I would have preferred to have a single layer schema first. We're still hammering out the details of the basic elements, but already we're opening new cans of worms left and right with the problems of describing multi-layer topologies.
Hi Jeroen
Outside of the domain this signal gets split, and both VLAN sets head of to their separate destination domains.
This is still unclear to me. How do you split Ethernet packets travelling on a single cable that are tagged with different vLANs? This can only occur in a Ethernet device (the Switching Matrixi in my example). I do not see how you do it otherwise. So your port needs to go to one other Ethernet port (portB in my example)
At this point you have splitted the vlans and proceed 'normally toward the to the two domains (via portC and portD in my example).
What you describe is what I meant with my third possibility, adding a virtual node.
The Switching Matrix is the virtual node, you mean?
PS. This problem, and many others like it is why I would have preferred to have a single layer schema first. We're still hammering out the details of the basic elements, but already we're opening new cans of worms left and right with the problems of describing multi-layer topologies. True:-)
Paola ps = in case we take this offline...
Paola Grosso wrote:
Hi Jeroen
Outside of the domain this signal gets split, and both VLAN sets head of to their separate destination domains.
This is still unclear to me. How do you split Ethernet packets travelling on a single cable that are tagged with different vLANs? This can only occur in a Ethernet device (the Switching Matrixi in my example). I do not see how you do it otherwise. So your port needs to go to one other Ethernet port (portB in my example)
My point is that the splitting of VLANs happens outside of your domain and outside of your control. So yes, it is going to happen somewhere in a switchingmatrix, but you cannot control it. So, how do you want to describe this? Do you want to introduce a Node/SwitchingMatrix that is there but you do not control? Jeroen.
Hi Jeroen Ok, I finally got your point.
My point is that the splitting of VLANs happens outside of your domain and outside of your control. So yes, it is going to happen somewhere in a switchingmatrix, but you cannot control it.
So, how do you want to describe this? Do you want to introduce a Node/SwitchingMatrix that is there but you do not control?
But isn't this the issue of topology exchange? And building a global topology from multiple sources. The SwitchinMatrix that does the splitting would be contained in the topology for domain B. The domainA would just describe its edge port, and the link to the other side. (With most likely the adaptations into Ethernet) So indeed you cannot introduce a Node that you do not control, because a priori you do not know of its existence. In a first moment, without having built a global view of the network you do not know what has happened to the vlan 1000-2000 and 3000-4000 that you just "threw over the fence". When a gloal topology gets constructed from the various sources consistency gets checked. The utlimate point I was trying to make was that in my opinion the schema you mailed us allows us to describe this usecase. Best regards, Paola
Hello, Paola Grosso wrote:
The SwitchinMatrix that does the splitting would be contained in the topology for domain B. The domainA would just describe its edge port, and the link to the other side. (With most likely the adaptations into Ethernet)
Yes, now I see where I made a mistake in my description, my apologies. The difficulty with this situation is that domain B handles the connection, and the splitting of the VLANs towards domains C and D. But domain B does not participate in the dynamic switching or topology description. In some ways this is similar to a trans-atlantic cable situation. You hand off the connection to a carrier in Amsterdam, and it comes out somewhere at the other end in MAN LAN, but you have no idea how exactly. Jeroen.
I apologize, my previous mail was more a less a repeat of what Paola had already said. Indeed, I think that the goal of NML is indeed to exchange topology information so that you know how/where the network traffic is going, even when part of the configuration is in another domain that you do not control. Of course, in this specific case, I would describe it as a static/fixed configuration, not as a (dynamic) SwitchMatrix (as you indicated that in this example the configuration is fixed) Regards, Freek
Jeroen van der Ham wrote:
- Describe a single port with two links attached to it. This means breaks the assumption that a port can have at most one bi-directional link attached to it.
- Describe a single port with two virtual ports each with its separate link. This breaks the assumption that a port is always connected by a Link (either a cross-connect or an external link), and we introduce a new concept of a virtual port. Are there other cases that would warrant this introduction?
- Describe a single port that is connected to a "virtual" node where the connection is split into two. This makes the description more complex, and also seems to allow a direct connection between the two inter-domain connections without using the original port.
Any other ideas?
I propose to describe at at two layers: - one port at the layer below Ethernet (physical link, if you will) which is connected to single a port in another domain. plus - two ports at the VLAN layer, each describing a logical connection on top of the underlying layer. and an adaptation to describe the relations between the two (sub)layers. This is how NDL is doing things, and it works pretty good. Regards, Freek Dijkstra
Freek Dijkstra wrote:
I propose to describe at at two layers: - one port at the layer below Ethernet (physical link, if you will) which is connected to single a port in another domain. plus - two ports at the VLAN layer, each describing a logical connection on top of the underlying layer. and an adaptation to describe the relations between the two (sub)layers.
This is how NDL is doing things, and it works pretty good.
How do you describe the fact that the separate VLAN sets go to different domains? You're only describing a single connection to another domain. Jeroen.
participants (3)
-
Freek Dijkstra -
Jeroen van der Ham -
Paola Grosso