|
Subject |
Draft minutes of joint session of Tech & Control Plane WGs |
|
From |
Kevin Meynell <kevin@xxxxxxxxx> |
|
Date |
Tue, 29 Nov 2005 22:25:49 +0000 |
GLOBAL LAMBDA INTEGRATED FACILITY TECHNICAL & CONTROL PLANE WORKING
GROUPS
Minutes of the joint session of the GLIF Technical Issues and Control
Plane Working Groups held on the 30th of September 2005 at Calit2, San
Diego, United States.
Kevin Meynell - Issue 1
PRESENT
Name Organisation Country
---- ------------ -------
Tomonori Aoyama U. Tokyo Japan
Bill St. Arnaud CANARIE Canada
Erik-Jan Bos (Co-Chair) SURFnet The Netherlands
Greg Cole GLORIAD United States
John Connolly U. Kentucky United States
Steve Corbato Internet2 United States
Steve Cotter Internet2 United States
Sergi Figuerola i2CAT Spain
Licia Florio TERENA -
David Foster CERN -
Pat Gary NASA/GSFC United States
Vikram Gazula U. Kentucky United States
John Graham UKLight United Kingdom
Eduard Grasa i2CAT Spain
Leon Gommans U. van Amsterdam The Netherlands
Jan Gruntorad CESNET Czech Republic
Jeroen van der Ham U. van Amsterdam The Netherlands
René Hatem (Co-Chair) CANARIE Canada
Michiaki Hayashi KDDI R&D Labs Japan
Akira Hirano UIC/EVL United States
Bonnie Hurst MCNC United States
Hideaki Imaizumi U. Tokyo Japan
Waturu Imajaka NTT Japan
Masahiko Jinno NTT Japan
Gigi Karmous-Edwards (Co-Chair) MCNC United States
Akira Kato WIDE Project Japan
Tomohiro Kudoh AIST Japan
Scott Macdonald e-side Japan
Joe Mambretti iCAIR/Northwestern U United States
Jun Matsukata NII Japan
Steve Meacham NSF United States
Kevin Meynell (Sec) TERENA -
Dan Nae Caltech United States
Naohide Nagatsu NTT Japan
Bram Peeters SURFnet The Netherlands
Ana Preston Internet2 United States
Jan Radil CESNET Czech Republic
Predrag Radulovic U. Tennessee United States
Ann Richeson Qwest United States
Paul Roberts U. Houston TLCZ United States
Yasunori Sameshima NTT Japan
Richard Schneider NASA/GSFC United States
Fay Sheu NCHC Taiwan
Matt Schmitz Cisco United States
Jerry Sobieski MAX/DRAGON/HOPI United States
Hideaki Tanaka KDDI R&D Labs Japan
Steven Thorpe MCNC United States
Christian Todorov Internet2 United States
Vasily Velikhov RRC Russia
Alan Verlo StarLight/TransLight United States
Steven Wallace Indiana University United States
Tom West National LambdaRail United States
Kennard White Glimmerglass United States
Garrut Yoshimi U. Hawaii United States
Oliver Yu UIC/EVL United States
1. WELCOME
Erik-Jan welcomed everyone to the meeting and introduced René Hatem
and Gigi Karmous-Edwards as the co-chairs of the session. The plan
was for the Technical and Control Plane Working Groups to meet
jointly during the morning session, then break-out into separate
groups in the afternoon.
The aim of the joint session was to discuss issues of relevance to
both working groups, as well as define the specific areas on which
each would work. This was necessary to divide-out the work and avoid
unnecessary duplication of effort.
After the various GLIF mailing lists were outlined, the question of
how to subscribe to them was raised. Kevin replied this was possible
from the individual working group web pages, but he would add links
to the GLIF home page (http://www.glif.is) to make things clearer.
ACTION 20050930-1: Kevin Meynell to make clearer on GLIF website,
how to subscribe to mailing lists.
2. DIVISION OF TASKS BETWEEN WORKING GROUPS
There was a discussion about the appropriate division of tasks
between the Technical and Control Plane Working Groups. René
suggested that the Technical Working Group should focus on compiling
information about GLIF resources, as it was necessary to find out
what was available where. It should also investigate automating
those lower-level tasks that currently required manual intervention.
Allied to this was the need to develop common service definitions,
although this should be done in conjunction with the Control Plane
Working Group.
? asked whether the proposed common service definition would be more
of an informal agreement, or whether it would be closer to an API.
Gigi replied that Jerry Sobieski planned to cover this in his
presentation during the separate Control Plane Working Group
session.
Joe suggested there first needed to be a standardisation of terms
within the GLIF community, otherwise there was the potential for
confusion. René replied this had already been discussed several
times, but there had been very little progress on the issue. Whilst
the issue had raised in many meetings over the past four years,
there had been very little discussion on the mailing lists where it
was easier to thrash-out such issues. However, in order to try and
progress the issue, the GLIF Secretariat was asked to start
collecting terms and definitions for a glossary.
ACTION 20050930-2: Licia Florio to collect terms and definitions for
glossary.
Matt added his observation that many issues were common to both the
Technical and Control Plane Working Groups, and it was important
they worked closely together. Erik-Jan replied this was one of the
reasons for having a joint session, and the plan was to continue
this in future.
3. GLIF RESOURCES
There were updates from several optical exchanges about their
current resources (see
http://www.glif.is/meetings/2005/tech/bos-resources.pdf).
Erik-Jan reported that NetherLight had 4 x 10 Gbps links to MANLAN,
2 x 10 Gbps to both StarLight and CERN, 1 x 10 Gbps to CESNET, GÉANT
and UKLight, 1 x 2.5 Gbps to NORDUnet, and multiple 10 Gbps to
SURFnet. Connected exchanges generally assumed bilateral control
over the links, although Internet2/DANTE controlled one of the
MANLAN links.
John reported that UKLight had 10 Gbps connections to both StarLight
and NetherLight. The extended development network within the UK was
based around four C-PoPs in London, Leeds, Reading and Warrington
connected via 10 Gbps links. There were ongoing 10 Gbps connections
from the C-PoPs to various institutions.
Lars reported that NorthernLight was a star network based around a
hub in Stockholm, with OC-48 connectivity to Copenhagen, Oslo,
Helsinki and NetherLight in Amsterdam. It is based on Cisco
ONS-15445s that provide 2 x 1 GE links between any two cities. It
will also be connected to GÉANT2 by December 2005, and will be
upgraded to n x 10 GE by June 2006.
David reported that CERN had 2 x 10 Gbps links to SURFnet, and 10 GB
links to both StarLight and MANLAN.
Christian reported that MANLAN had five 10 GE links to Abilene,
ESnet, GÉANT, NYSERNET and SINET. There were also five 1 GE links to
CA*net, SURFnet, Egypt and Qatar. Further OC-192 links existed to
Abilene, CA*Net (x2), GÉANT, SURFnet (x2), and a CANARIE relay to
Amsterdam.
René reported that CANARIE-StarLight had seven OC-192 links to
Toronto (x2), Amsterdam, Winnipeg, Vancouver, F10, and the GLIF HDXc
node. A further two OC-192 links would shortly connect IRNC and
Force10 (2nd connection). There was also an OC-48 link to TWAREN via
the GLIF node.
René reported that the CANARIE-Pacific Northwest GigaPoP had five
OC-192 links to Calgary, Vancouver (x2), Victoria, and the GLIF HDXc
node. A further two OC-192 links would shortly connect Busan (South
Korea) and IEEAF Tokyo. There were onward 1 GE connections from the
GLIF node to Pacific Wave and KREONET.
Akira reported that T-LEX had OC-192 and OC-12 links to the Pacific
Northwest GigaPoP. There were also 1 GE links to APAN, ASCC
(Taiwan), Data Reservoir, JGN2, WIDE, and the University of Tokyo. A
further OC-192 to IEEAF was planned in the future.
? reported that KRLight had 2 x 10 GE links to KREONET, and OC-192
links to both Hong Kong (HKLight) and Seattle (Pacific Northwest
GigaPoP). This provided onward connections to Beijing, Novosibirsk,
Moscow and Amsterdam.
? reported that UltraLight was comprised of six PoPs in Chicago
(StarLight), Geneva (CERN), Los Angeles (CENIC), New York (MANLAN),
Seattle (National LambdaRail), and Sunnyvale (National LambdaRail)
interconnected by 10 Gbps links. There were also external
connections to CANARIE, NetherLight and UKLight.
Erik-Jan suggested that it would be useful to list the various GLIF
resources on the GLIF website. Kevin said he would put a page
together from the presentations, but hopefully this information
could be extended.
ACTION 20050930-3: Kevin Meynell to list GLIF resources on GLIF
website.
4. GLIF REPOSITORY
Erik-Jan said that a more standardised system of representing GLIF
resources was required. As existing networks grew and others
appeared, it would be increasingly difficult to coordinate all the
lightpaths, interfaces, exchange points, policies and administrative
information in the current ad-hoc fashion. Although the aim was to
initially focus on the lower-level services, a coordinated
repository of information could also be useful for the control plane
levels as well.
At the present time, there appeared to be three main approaches to
implementing network resource repositories. GLORIAD had used a
monolithic database approach for their purposes, whilst NLR had
adapted the DNS to store network information. The third, but thus
far more theoretical approach, was to utilise semantic web
technology to create a distributed database of network information
that could be maintained by each organisation. The merits of each of
these approaches needed to examined to determined whether a common
standardised system could be agreed upon.
4.1 Using RDF to describe networks
Jeroen gave a presentation about how the Resource Description
Framework (RDF) could be used to describe GLIF facilities (see
http://www.glif.is/meetings/2005/tech/vdham-rdf.pdf). Network
descriptions were necessary to provide an overview of resources,
make path discovery easier, and to undertake simple problem
detection, but at the present time, GLIF resources were coordinated
through an ad-hoc system of scribbled diagrams and telephone/e-mail
communication. Such descriptions needed to be readable by both
humans and computers, and semantic web techniques offered this
possibility.
The semantic web is an extension of the current web in which
information is given well-defined meaning, with RDF being a
lightweight ontological framework for undertaking this. It describes
things using triplets consisting of subject (the thing being
described), predicate (a property of the thing being described) and
object (the value of the property) elements. A unique terminology is
achieved using namespaces based on Dublin Core which are both human
and machine readable.
A Network Description Language (NDL) based on RDF has been devised
by UvA. This allows resource description files to be created, which
can then be queried using an SQL-like language such as SPARQL. Such
descriptions can also point to other network descriptions, allowing
a complete picture of resources to be built-up. Another way of
querying remote information is to issue SPARQL queries via HTTP or
SOAP.
With respect to GLIF, each organisation can create a NDL description
of their network, which could be validated and published via a
portal which users would use to find paths and resources. This
information could also be linked to geographical references (e.g.
Google Earth) to provide mapping functionality, as well as
administrative and policy information.
The next step is to develop tools that automatically generate and
update descriptions. In addition, the portal should be developed to
link these descriptions together to provide a comprehensive overview
of GLIF resources. NDL could also be extend to support higher-level
protocols, as well as security and filtering.
Jerry asked how quickly information was updated using this system.
Jeroen replied that institutions published this information
themselves, and it could be updated in real-time. It was therefore
faster than a monolithic database approach.
4.2 GLORIAD Database Approach
Greg Cole gave a presentation about GLOCIS (see ?).
Cees commented that it was important to ensure network information
was up-to-date, and asked whether there was a web interface where
this information could be entered. Greg replied this was planned,
but had not yet been implemented.
Cees also asked whether there were plans to extend the system and
incorporate other types of information. Greg replied they wanted to
use the system for at least six partner countries.
Gigi asked whether the mapping components would be released. Greg
replied they planned to release the source code shortly.
4.3 Mapping US Regional Networks
Steven gave a presentation on using the DNS to map GLIF resources
(see http://www.glif.is/meetings/2005/tech/wallace-dns.pdf). He had
estimated that 55% of optical networking resources in the US were
outside of National LambdaRail, and so an organised system of
coordination was becoming increasingly important. The Quilt had been
established as an informal organisation for purchasing bandwidth,
knowledge exchange, and sharing of test equipment, but it did not
have a good overview of regional network resources. In March 2005,
it was therefore agreed this should be mapped, initially using hand-
collected data. However, this was felt to be somewhat cumbersome, so
ways of automating the process were investigated.
The DNS was chosen as the mechanism for this, as it was a well-known
and widely-established distributed database that offered user-
defined record types. The domain 'network-map.net' was therefore
registered, and a record structure was defined. A virtual
cartographer program was also developed that was able to query the
DNS and make maps on demand.
Steven was interested in extending the system and asked who would be
willing to maintain their own zone files. His lab would be willing
to continue to host 'network-map.net', populate zone files for the
existing US RONs (Regional Optical Networks), and make the virtual
cartographer software available.
Cees thought the system would make too heavy use of the DNS, and
neither was it designed for this purpose. Too much information would
need to be included, and it would not be sufficiently real-time.
However, it might be possible to utilise SRV records to point to
locations where resource information could be found.
John asked what DNS offered that something like LDAP did not. Steven
replied the DNS offered a much more simple delegation mechanism.
A quick show of hands revealed that there did not seem to be a great
deal of support for utilising the DNS in this manner. Nevertheless,
it was felt the experiences of the working system could be used to
help refine the data model. Furthermore, there was interest in
utilising the virtual mapper.
Summary
Erik-Jan concluded that the different approaches served particular
purposes and were not essentially in competition with each another.
In many respects, the various activities could be complementary to
each other if the community could devote some effort to combining
them. There was generally a lot of support for the RDF model, whilst
GLORIAD had developed some very nice mapping capabilities. At the
same time, the DNS could be used to point to where network resource
information could be found.
It was agreed that a sub-working group comprising Erik-Jan Bos,
Greg Cole, Freek Dijkstra, Lars Fischer, Jeroen van der Ham and
Steven Wallace should be established to investigate how the various
approaches could be combined. The GLIF Secretariat was asked to set-
up a mailing list to facilitate this.
ACTION 20050930-4: Kevin Meynell to set-up mailing list for
repository sub-working group.
5. NEXT MEETINGS
The 6th Annual Global Lambda Workshop would be held on 11-14
September 2006 in Tokyo, Japan. It was anticipated this would
include meetings of both the Technical and Control Plane Working
Groups.
It was also agreed that interim meetings of both working groups
should be held in early-2006. The working group chairs in
conjunction with the GLIF Secretariat would investigate suitable
dates and venues for this.
ACTION 20050930-5: Technical and Control Plane Working Group Chairs
to investigate suitable dates and venues for interim meetings.
[It was subsequently agreed that the interim working group meetings
would be co-located with the Internet2 Joint Techs Workshop in
Albuquerque, USA; probably on 8-9 February 2006.]
OPEN ACTIONS
20050930-1 Kevin Meynell to make clearer on GLIF website,
how to subscribe to mailing lists.
20050930-2 Licia Florio to collect terms and definitions for
glossary.
20050930-3 Kevin Meynell to list GLIF resources on GLIF website.
20050930-4 Kevin Meynell to set-up mailing list for
repository sub-working group.
20050930-5 Technical and Control Plane Working Group Chairs
to investigate suitable dates and venues for interim
meetings.
