[Nml-wg] OGF 23 discussions: pictures
Freek Dijkstra
fdijkstr at science.uva.nl
Thu Jun 5 12:01:58 CDT 2008
Hi all,
OGF was very useful, at least to me, and a LOT of papers notes were
scribbled scribbled in discussions after the session.
I took the liberty of recreating an electronic copy of a few of these
pictures. You will find them attached.
Slide 1: Node Virtualization
----------------------------
During the session, we had a discussion about physical versus virtual
nodes. Later, it turned out we had to distinguish two different concepts:
a) partitioning of devices into multiple logical devices. Typically,
control of each logical part will be given to another party.
b) abstraction of multiple device and expose it as one single device
(for example, Raptor switches and multi-chassis routers do this).
In the picture, the green boxes are physical devices -- real hardware, a
chassis. The blue boxes in the physical devices represent logical
partitionings, referred to as "virtual nodes".
Note: We first used the terms "nodes" for logical nodes and "device" for
physical devices. Later we agreed that "node" and "device" should be
interchangeable, and we should refer to "physical nodes" and "virtual
nodes" (or "physical device" and "virtual device"). The parent concept
is called "node".
A group of nodes (either virtual nodes or physical nodes) can be exposed
as an abstracted thing that behaves as if it was a device (with
interfaces, capabilities, etc.). This is a "network", or "graph", since
it must behave as a connected graph (thus there is a implicit assumption
that data can be transported between any two interfaces of the same
graph/network.
Slides 2+3: Node Relations
--------------------------
This slide gives the relationship between node and network, and between
virtual and physical nodes. Each node has an attributed "implemented".
If this is undefined, it is a physical device. If it is set, it means
that this is a virtual node, which is a partition of the referred instance.
Note: this schema only gives the relations between classes (one to many,
many to one). In this example, the "implemented" attributed is a many to
one attribute; the others are one to many. To convert this to a
relational database, it is much more convenient to use many to one than
one to many attributes. I suggest we follow that practice in the final
schema.
Slides 3 is a variant of slides 2. In here, virtual node and physical
node are subclasses of the base node class. This is useful if one of
them has attributes that the other does not. I put location in there,
but that may be a bad example.
The network/graph abstraction has interfaces, just like regular nodes.
These can be inferred (derived) from the original nodes in the network.
Therefor, this is not an attribute, but a function that calculate the
edge interfaces of the network. Of course, in (XML or RDF) messages,
this relation is explicit -- it just is implied in a database (to avoid
internal inconsistencies).
Slide 4: Networks
-----------------
There are two concepts of the term "network". Martin Swany noted that in
the IP world, the word "network" has a very well defined meaning, the
"link local" network between routers.
The definition which was proposed in the session by Pascale Primet was
that of a connected graph. This is what is meant by "network" when
people talk about e.g. the "Internet2 network" or "the SURFnet network".
John Vollbrecht proposed to use the term "multi-point link" for a
IP-terminology "network", and use the term "network" for the graph-like
"network".
This image attempts to show the two terms in a single picture. While the
definition was agreed upon, the term was not. We may use "G-Network"
(for graph-like network) and "I-Network" (for IP-term network) until we
decide on this.
Slide 5+6+7: G.805 Functional Elements
-----------==-------------------------
We had a discussion on the terms "link", "path" and "connection". In
particular, it should be noted that a path on a lower layer is a link on
a higher layer. For example, an IP "link" may in fact cross a while DWDM
network, or Ethernet network. G.805 defines exactly this. This slide
re-iterates the terminology of G.805.
Slide 6 relates this to "our" terminology:
- link = G.805 link connection
- path = G.805 network connection
- circuit = G.805 trail
A trail is a terminated network connections (thus with error
correction, retransmission, etc.). John Vollbrecht likes the distinction
between these two terms, but there was no consensus for that.
Slide 7 is a picture I added to explain G.805 for those really
interested (including subnetworks and multiplexing adaptation).
Ask me if you are interested.
Groups
------
(not in the slide pictures)
We agreed that we should keep the concept of group, and that path and
domain, and perhaps also network are specific types of groups.
Function/Service
----------------
Martin Swany argued that adaptations and cross connects are functions or
services in a node. This was not extensively discussed in the group
itself, but in general the idea was met in favour.
Note: it is not clearly defined yet what a service or function is. For
example, if it is a transport function between ports, this includes
adaptations, cross connects and link connections. An alternative is to
reserve the term only for operations within a node (thus excluding link
connections).
Regards,
Freek Dijkstra
-------------- next part --------------
A non-text attachment was scrubbed...
Name: Modelling_Notes.ppt
Type: application/mspowerpoint
Size: 499712 bytes
Desc: not available
Url : http://www.ogf.org/pipermail/nml-wg/attachments/20080605/f24b7e87/attachment-0001.bin
More information about the nml-wg
mailing list