Network modeling & NML group in the OGF
Freek Dijkstra <fdijkstr <at> science.uva.nl>
2008-05-07 14:49:55 GMT
I've was recently pointed to the WSON list by a colleague, who is
liaison of the Open Grid Forum (OGF) towards the IETF .
The list has no archive, so I may have missed a few topics, but after
reading the discussion on the difference between OOO and OEO or on
wavelength versus waveband switching, I think there is a need for an
explicit multilayer model. The MLN/MRN work at least has a syntax, but I
have not seen a formal model.
I'll post this to the CCAMP list rather than the WSON list since it is a
bit more general.
I just finished my PhD thesis on multi-layer path finding. I'll briefly
mention this work (I apologize for the shameless plug) and than mention
was is currently done in the OGF. Let's see if there is some mutual
We created a model and syntax called Network Description Language (NDL)
. A major goal was to enable complex multi-layer path finding, with a
technology independent algorithm. "Complex" because multi-layer path can
in fact contain loops , which is not possible in single layer networks.
The technology independence (no technology-specific code) means that not
only the networks are described in an abstract model, but so are the
different technologies. For example, the path finding "learns" about
characteristics of WDM by reading a file describing the technology. That
file contains the different adaptations, layers and label definition (in
our WDM example file, the labels are floats: the wavelength in nm, and
the spacings are described as different adaptations. Interestingly it is
possible to use the same model to make another technology description of
WDM, for example with or without a group mux sublayer). Of course the
challenge was to force the different technologies in such a generic
model. Tagged Ethernet in particular took some effort. One thing that
worked very well was to distinguish between two different switching
capabilities: one "switching" for forwarding traffic with the same
"label" (e.g. wavelength switching without wavelength conversion or
Ethernet VLANs) and for forwarding while allowing label changes (e.g.
timeslots in TDM or wavelength conversion).
Beside the model, we create a syntax. Instead, our syntax is based on
semantic web  and RDF. What worked very well was that we could easily
separate the different subtopics: topologies description, (technology)
layers, capabilities (such as the ability or inability of a device to
convert between labels), domain abstraction, physical properties, etc.
What's more, we could easily combine this information with different
information such as policy descriptions and pointers to reservation
systems. We are in the process of combining network description with a
description of 4K very high definition digital cinema display, media and
storage) , so that a metascheduler not only can take network
resources but also storage and visualization resources into account.
In case you wonder how our model compares with GMPLS: Jeroen van der
Ham, a fellow PhD student, just wrote an interface to convert between
our description and OSPF-TE. This was done in collaboration with Max
Gigapop, a research network that also developed DRAGON, a free GMPLS
Our work is developed and in limited use in a community of national
research networks called the GLIF , which provides circuit-switched
network connection for e-science applications around the globe.
("limited" since it is restricted to a single layer for now).
As for the relation with GMPLS: our model is based on functional
elements as described in ITU-T G.805 , combined with the label
concept in GMPLS. That is generic and powerful enough to describe and
find paths in WDM, TDM, Ethernet and fiber switch networks .
So much for our model and the shameless plug.
The reason I mention all this is that we are of course not the only ones
to create a syntax to describe networks. OSPF-TE is an obvious other
one. There are quite a few research projects which also create their own
syntax. Some use GMPLS, though most that involve multi-layer networks
create their own model and syntax. About a year ago, some of these
groups joined forces to standardize on a network description.
The Open Grid Forum (OGF) , a standardization body originally active
in standardization of Grid software, has a working group, called Network
Markup Language (NML), which does exactly such multi-layer network
modelling . Active members are us, developers of the PerfSONAR
measurement software, and miscellaneous partners in GÉANT (European
research network) projects. Each of these active members brings their
own multi-layer models, each with each strengths and weaknesses.
Clearly there is some overlap with CCAMP here. Of course, there are
difference as well. The NML working groups mostly is concerned with
multi-domain descriptions, and the interactions needed there. Thus, the
ability to describe an abstracted view of the network, with capability
information, pointers to services for policy decision and policy
enforcement points, etc. Typical projects I see are research projects
that provide dedicated network connections (either using Ethernet, DWDM,
or TDM) through a collaboration of 5-20 network domains, each using
their own inter-domain control plane software. Typical control planes I
see are DRAGON, UCLP, G-Lambda, DRAC, AutoBAHN, etc.
If you are interested, I gladly elaborate on either our model, the path
finding, or the working group.
Perhaps it would be could that we can share some of our use cases,
requirements and solutions, while you some of the active members here
could provide the NML working group with input from the CCAMP group.
With kind regards,
 OGF http://www.ogf.org/
 Network Description Language,
 Loop example in multi layer networks
 Semantic web, http://www.w3.org/2001/sw/
 4k digital cinema, http://www.cinegrid.org/
 DRAGON, http://dragon.maxgigapop.net/
 GLIF community, http://www.glif.is/
 ITU-T G.805, http://www.itu.int/rec/T-REC-G.805/
 Technology descriptions in NDL,
 NML-WG, https://forge.gridforum.org/sf/projects/nml-wg