A rather long sequence of messages - so lists can see the discussion. John Begin forwarded message:
From: Martin Swany <swany@cis.udel.edu> Date: May 12, 2009 9:24:46 AM GMT-04:00 To: Jeroen van der Ham <vdham@science.uva.nl> Cc: John Vollbrecht <jrv@internet2.edu>, Freek Dijkstra <fdijkstr@science.uva.nl
, Aaron Brown <aaron@internet2.edu>, Jeff Boote <boote@internet2.edu>, Andrew Lake <alake@internet2.edu>, Paola Grosso <grosso@science.uva.nl>, Cees de Laat <delaat@uva.nl>, Richard Hughes-Jones <richard.hughes-jones@dante.org.uk>, Guy Roberts <guy.roberts@dante.org.uk> Subject: Re: NSI naming
Hi Jeroen,
I'm (re-?) adding Jason to the cc list as he's right in the middle of this discussion as well.
I would add that I think that we are all very close on the way in which things are related, which gives rise to layers/levels of the network. I think that the "Relation" element that we're using in UNIS (perfSONAR/DCN) land really makes the expressions there start to look like the RDF 3-tuples in a way. While that problem is hard, I think that we've been inching closer to having a common solution framework.
best, martin
On May 12, 2009, at 4:13 AM, Jeroen van der Ham wrote:
Hi,
I think Martin's analysis of your description is correct. I only want to add that you seem to be contradicting yourself regarding layering, and I think it is very important that we clear that up.
Before I begin, I want to clear something up: Networks transport data by encapsulating it in different layers. I think we can all agree on that analysis, and we all know what layers are in general. Therefore I don't want to complicate this (already complex) discussion by introducing yet another term (i.e. "level").
Then on to your analysis: in the beginning you say that NSI wants to describe links at a certain layer going over links at a lower layer. Then later on you say that NSI operates at a single layer for a particular request.
Current networks operate at multiple layers (Ethernet, SONET, Lambda, etc.). Therefore, if the NSI wants to be able to provision services across these networks, then it will have to handle multiple layers.
Please realise that the NML is currently describing a layer-less model. Layers are not part of the model and we are only describing the general things that are part of every layer. So NML is currently not able to describe that links at one layer run over a link at a lower layer.
As Martin said, describing layers, and the adaptations between them is a very very hard problem. One that the NML group will be taking on soon though.
Jeroen.
--- For, reference, we have updated the terminology spreadsheet here: http://forge.gridforum.org/sf/go/doc15512?nav=1 I left out the cases that were OK.
This is a link at a level. This seems a fine name. For NSI the question is if the link can carry links at a different level. I think so - I think there is a relation which says a link at one layer can support multiple links at a different layer. Examples would be multiple VLANS over ethernet or multiple VCGs over SONET, or multiple waves over a fiber.
This is definitely part of the model.
<pastedGraphic.pdf> logical port (per layer) - G.805 calls this a connection point, NML and NMwg call it a port. In NSI we started discussion calling it a Port but switched to edge point to avoid the physical connotations of port. The groups should agree on this name or include an alias for different names if we can't agree.
I think that the NML has agreed to call this a port. Our naming was changed a while ago (due to ITU naming!) in perfSONAR and I believe the NDL group has agreed. Certainly any term carries connotations, but the pervasive opinion is that they are more similar than they are different and it is useful (for pathfinding) to use the same underlying term, but specialize it for different purposes.
<pastedGraphic.pdf> concatenated series of links - Path in NML and NMwg, concatenated series of links in G.805. It seems to me that this should be a link not a path. I am not sure when some thing is a path rather than a link. I also think a concated series of links is a link (maybe it needs a name to become a link?)
In my mind, there is often a one-to-one relationship between paths and links and the only question as to which you use is whether you need to see the internals or not. I think this addresses and issue below as well.
<pastedGraphic.pdf> Network Layer (topology on a single layer) - This seems an important concept - topology on a layer. A couple parts of this that I think may need definition: 1) how to tie together administrative domains with topology - for example one administrative domain may know about a subset of what another administrative domain knows; and 2) How do layers interact - is one layer carried by or used by another?
<pastedGraphic.pdf> aggregated device -- called topology in NML. this might be a network in NSI, but I am not sure that fits well.
I don't know that there's agreement that an aggregated device is a topology, but it sounds like this isn't an issue
<pastedGraphic.pdf> domain - this seems like a network in NSI terms. Certainly it has a lot of the same characteristics. But it does not have all the characteristics - see discussion on "what is missing" below
The easiest answer is that we can add characteristics and plan to do just that as the base elements are "subclassed" into specific uses. But I do understand that there is a little gray-ness between Networks and Domains. My own take is basically this: Networks are addressing scoped, and Domains are administratively scoped. I take that largely from Internet terminology and I know there are counter- examples. So, the more complicated answer is that there are a number of different kinds of groups and we need to see which ones fit.
What seems missing --
1) The description of adapatation in general: for example in a concatenated series of links, each concatenated link may be carried on a different sort of higer layer link. A VLAN might be carried on an Ethernet link (as a VLAN), on a SONET link as GFP encoded VLAN. In this case the underlying link with its coding is carried over different higher layer links with their coding.
I believe that this is a good example of where the Link/ Path dichotomy helps. If we need to know that details of of the different sorts of constituents, we need to access the path object, otherwise it's just a link. Also note that I don't really think we have any fixed notion of "higher" and "lower" layers now. I know that we did in perfSONAR but there were so many special cases of relationships (L2TP, q in q, ip in ip) that we pretty much gave up. So, I don't think there are any problems with representing what you're talking about above, so maybe we should try to look at specific examples?
2) The concept of connection oriented network: A NSI network is something that can create segments (aka links a lower layer) between edge points (aka ports). A network participates with other networks in a topology where the elements are networks that connected with links. In graph theory network would be node, and link an edge.
I think that a Link represents a network connection pretty well. Maybe the Path/Link discussion is convincing, but I should say that's more of the perfSONAR/DCN perspective and not that of NDL (necessarily -- I think it's close, but I won't speak for them.)
3) Names for the elements within a layer that are concatenated. NSI calls these segments for now. The NSI topology elements are links and networks. Each of these elements can provide segments and the segments can be concatenated to create a ete segment. I believe that segments and links at a level are identical concepts, so perhaps we can come up with an alias for how to describe NSI segments and ete concatenated segments.
Again, hopefully, the Path/Link concept describes this for the complicated case. (And a simple relation works when 4K VLAN links are carried over the same Link....) best, martin Hi Freek - Thanks for the comments. I did mix NML and NMwg terminology. Sorry about that, I have not been sure what the NML terminology is in a lot of cases. I am now looking at the spreadsheet you mentioned - which I attach so we are sure we are talking about the same thing. I think this is a great tool for discussing the issues. As noted at the end of the note I suggest we have a call prior to OGF to talk about this and that we try to sort these issues out at OGF. I think we are close, but I at least have some questions and issues as discussed below. I will go down the sheet with comments and what I think the item might be called in NSI. The NSI naming intends to use NML names where the underlying concepts are the same. I was trying to put these into the spreadsheet, but it doesn't seem to fit in a good way so I will just comment and try to figure out how to present it better later. This is a link at a level. This seems a fine name. For NSI the question is if the link can carry links at a different level. I think so - I think there is a relation which says a link at one layer can support multiple links at a different layer. Examples would be multiple VLANS over ethernet or multiple VCGs over SONET, or multiple waves over a fiber. logical port (per layer) - G.805 calls this a connection point, NML and NMwg call it a port. In NSI we started discussion calling it a Port but switched to edge point to avoid the physical connotations of port. The groups should agree on this name or include an alias for different names if we can't agree. End point of a connection. seems a logical port with links connected to it. I think not a new concept for NSI concatenated series of links - Path in NML and NMwg, concatenated series of links in G.805. It seems to me that this should be a link not a path. I am not sure when some thing is a path rather than a link. I also think a concated series of links is a link (maybe it needs a name to become a link?) end-to-end connection - this seems a concatenated link with links to ? I skip monitored and multipoint connection/link Network Layer (topology on a single layer) - This seems an important concept - topology on a layer. A couple parts of this that I think may need definition: 1) how to tie together administrative domains with topology - for example one administrative domain may know about a subset of what another administrative domain knows; and 2) How do layers interact - is one layer carried by or used by another? I skip partitioned device - I am not sure how it would apply for NSI aggregated device -- called topology in NML. this might be a network in NSI, but I am not sure that fits well. domain - this seems like a network in NSI terms. Certainly it has a lot of the same characteristics. But it does not have all the characteristics - see discussion on "what is missing" below I skip encapsulation of data and addition of monitoring - I do think these are adaptations. Network layer -- This seems to go with the concept of link at a layer. I think that NSI operates at a single layer for a particular request. Data encoding of a network layer. I would like to see examples of what this means. For example does it mean GFP coding of VLANS over VCGs? Also, I have a question about whether these definitions apply to data`only - in NSI discussion has been that connection capabilities should support any technology, including for example, waves carried over fiber - where waves are switched without going to digital. Physical Link - this seems a link at a layer - not special for NSI, but important for others I skip physical Interface and device Switching capability - I am not sure this is needed in NSI - see discussion below Cross connect -- this is what NSI does at edge points - connects segments from different sources to concatenate them Channel over a link - A channel seems like a NSI segment. It could be described as a link over a link at a different layer. What seems missing -- 1) The description of adapatation in general: for example in a concatenated series of links, each concatenated link may be carried on a different sort of higer layer link. A VLAN might be carried on an Ethernet link (as a VLAN), on a SONET link as GFP encoded VLAN. In this case the underlying link with its coding is carried over different higher layer links with their coding. 2) The concept of connection oriented network: A NSI network is something that can create segments (aka links a lower layer) between edge points (aka ports). A network participates with other networks in a topology where the elements are networks that connected with links. In graph theory network would be node, and link an edge. 3) Names for the elements within a layer that are concatenated. NSI calls these segments for now. The NSI topology elements are links and networks. Each of these elements can provide segments and the segments can be concatenated to create a ete segment. I believe that segments and links at a level are identical concepts, so perhaps we can come up with an alias for how to describe NSI segments and ete concatenated segments. Are there questions or corrections? Comments on how to come to understanding and consensus for this? I suggest that we might want to have a call to start discussions on this before OGF and to try to resolve issues at OGF. There is a NSI call every Wed at 9ET - we could have a joint call on Wed May 20 if that seems reasonable. John On May 7, 2009, at 10:21 AM, Freek Dijkstra wrote:
John Vollbrecht wrote:
Attached is a try at defining needs of NSI from NML. Some if it uses NMwg terminology as I understand it. I am thinking it would be good to sort this out and get a good definition by the end of OGF.
Hi John,
Thanks for writing this down -- a few comments.
You mix the NMwg with the NML terminology -- the NML terminology has improved much since the NMwg documents, and will deprecate the NMwg schema.
Here is a comparison table between NML and NSI, as far as we see. NSI NML End Device --- Node Edgepoint --- Port Network --- Topology (*1) Link --- Link Segment --- (under discussion) Service Group --- n/a Level --- Layer
(*1) If I understand your "network" correctly; the term network has different meanings.
Segment is still under discussion. We've used the term "Channel" a few times, but there is no group consensus. Segment refers to a multi layer concept, which will be defines in a separate schema.
NML hasn't defined Service Group. It seems an important concept, so perhaps we should.
Have you seen the terminology comparison table at http://forge.gridforum.org/sf/go/doc15512
Regards, Freek Dijkstra Jeroen van der Ham
John Vollbrecht Senior Network Engineer, Internet2 office 734 352 4960 cell 734 395 7890