[Nml-wg] Discussions at OGF 21

[Freek Dijkstra]

For those who haven't been at the OGF, there have been two NML-WG
sessions at OGF 21:
http://www.ogf.org/gf/event_schedule/index.php?id=894
http://www.ogf.org/gf/event_schedule/index.php?id=895

Below are my personal notes (these are NOT the meeting notes!!) I wrote
it for someone else and my own reference, but mail them to the list, as
they may be useful to others.

As you can see, the first day consisted of 4 talks of 10 minutes, which
together took all 90 minutes of the session. :-)

The Tuesday session was for discussions. My goal was to define a list of
"buckets" (modules or schemas that are (mostly) independent subtopics of
the larger concept of "network descriptions"), and decide which
subtopics should NOT be part of NML, effectively limiting the scope of
the working group.

See my mail of May for a list of "buckets":
http://www.ogf.org/pipermail/nml-wg/2007-May/000005.html
In addition, Anand Patil suggested a new subtopic, I hadn't thought of
yet: time-based information.

However, we didn't do this. Instead, we defined use cases and topics,
which was also a very fruitful discussion. Below, I'll quote the topics
we came up with.

In addition to this discussion, we made another in my opinion very
import decision: ask each creator of an existing schema (e.g. at least
NM-WG schema, cNIS and NDL, and perhaps the schema used by Grid5000 as
well) and model that in UML, and give that as input to the NML
workgroup. This was already suggested by John Vollbrecht on the list,
and repeated at the session by Martin Swany, and I wholeheartedly agree
that this is a good way forward: with this, we can easily see the
commonalities between the existing work, and use that as a starting point.



Use Cases
---------
  Use cases are example applications. NML should meet the
  requirements of these apps.
  This list was compiled at a discussion at OGF 21.
- perfSonar: network measurement
    - needs concept of end points and actual network connection
      in between.
    - description of the entire network connection (the
      tandem of individual link connections)
- fault isolation
    - different levels of abstraction
    - pointers to information services with
      detailed/other/more information
    - filters showing subset of entire network (e.g.
      only allowed/available resources)
- bandwidth on demand service
    - needs concept of end points and potential
      network connection in between
    - need time information, start time, end time, ...
    - link-characteristic:  bandwidth, jitter, etc.
    - concept of "variable" end point (any cluster
          meeting other requirements)
- path finding
    - (PCE in GMPLS terminology), need to specify link
      weight
      - is the path computation element part of NML
        or would it use NML?
    - need hooks to authorisation, policy (including
      device/configuration based policies),
      reservation/usage database
        - do we need pointers from NML to auth/usage,
          pointers from auth/usage software to NML, or
          include auth/usage information in NML?
    - description of the entire network connection (the
      tandem of individual link connections)
- domain abstraction
- inventory
- shared links (both conflict of resources; shared risk
  links?)
- multilayer (layer 2, layer 3 characteristics for a
  link)
- logical framework
    - links in a network connection are transitive
    - semantical framework to allow logical reasoning
      on network elements

Existing Work
-------------
- Network Description Language (NDL)
    RDF Schema by the System and Network Engineering group of the
    University of Amsterdam.
- the UML schema used in perfSONAR for network description
- Network descriptions used in Grid5000
- the UML schema used in the cNIS project in GÉANT 2 for network
  description
- AutoBAHN stitching framework
    See GÉANT2-07-066v5 - Deliverable 3.5.3 -
    Report on Testing of Technology Stitching.pdf

Topics
------
- visualization
- path finding
- network inventory
- network measurements

Buckets
-------
  Buckets are more or less separate topics. Ideally, there
  should be few dependencies between the different
  buckets/topics/subjects/umbrella/schemas/ontologies.
- Topology (single layer)
  e.g. devices, (virtual) interfaces, links, network connections.
  Re-use existing terminology? e.g. ITU-T G.805, Graph theory,
  GMPLS, ...
- Physical equipment
  e.g. physical devices, locations, geographic location
- Organisational boundaries
  e.g. domains, owner domain, administrative domain,
  organization/people
- Device/domain capabilities
  can switch (at which layer), interworking, possible labels
- Device configuration
  internal link connections, available labels
- End-to-end network connection
  usage (path setup request, reservation ID, specific path, hops)
- Time-based data
  Current status, snapshot or time-based, change management, ...
- Technology characteristics
  Thus meta-data to describe a technology, not technologies itself.
  e.g. layer, technology, adaptation
- Circuit switched properties
  e.g. labels, channels, capacity
- Packet switched properties
  e.g. source & destination labels, buffers, packet size, bandwidth
- Layer 1 specifics
  e.g. TDM, WDM, Fiber, UTP, wireless, Fiber bundles (ducts), OTN, ...
- Layer 2 specifics
  e.g. Ethernet (MAC + VLAN), ATM, PPP, L2TP, Myrinet/Infiniband, ...
- Layer 3+ specifics
  e.g. IP, TCP, ...
- Policies
  Authentication, authorisation, ...
- Scheduling
  Reservation, ....

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 249 bytes
Desc: OpenPGP digital signature
Url : http://www.ogf.org/pipermail/nml-wg/attachments/20071018/11a64ea6/attachment.bin