Draft minutes of the 4th TF-TANT meeting (issue 2)
|
|
|
TERENA/DANTE TASK FORCE FOR TESTING ADVANCED NETWORKING TECHNOLOGIES
Minutes of the 4th TF-TANT meeting held on the 17th and 18th of June
1999 at INFN-CNAF, Bologna, Italy.
Kevin Meynell - Issue 2
PRESENT
Name Organisation Country
---- ------------ -------
Werner Almesberger EPFL Switzerland
Michael Behringer Cisco United Kingdom
Marc van den Bergh KPN Research The Netherlands
Marjan Bogatinovski KPN Research The Netherlands
Zlatica Cekro VUB/ULB Belgium
Phil Chimento Uni. Twente The Netherlands
Howard Davies DANTE -
Tiziana Ferrari INFN Bologna Italy
Silvia Giordano EPFL Switzerland
Leon Gommans U.Utrecht/Cabletron The Netherlands
Ivano Guardini CSELT Italy
Christoph Graf (Chair) DANTE -
Mark Jansen Uni. Utrecht The Netherlands
Joop Joosten CERN Switzerland
Tom Kosnar CESNET Czech Republic
Simon Leinen SWITCH Switzerland
Silvia Matteoni Uni. Bologna Italy
Kevin Meynell (Sec) TERENA -
Paolo Moroni CERN Switzerland
Jan Novak DANTE -
Vaclav Novak CESNET Czech Republic
Simon Nybroe Telebit Denmark
Giovanni Pau Uni. Bologna Italy
Antonio Pinizzotto IAT-CNR Italy
Esther Robles RedIRIS Spain
Gianni Rossi CSELT Italy
Roberto Sabatino DANTE -
Guenther Schmittner JKU/ACOnet Austria
Jeremy Sharp UKERNA United Kingdom
Robert Stoy RUS/DFN Germany
Jean-Marc Uze RENATER France
Christina Vistoli INFN Bologna Italy
Bert Wijnen IBM The Netherlands
Wilfried Woeber ACOnet Austria
Apologies were received from:
Mauro Campanella INFN Milano Italy
Olav Kvittem Uninett Norway
Cees de Laat U.Utrecht The Netherlands
Ladislav Lhotka CESNET/USB Czech Republic
Vassilis Merekoulias GRNET Greece
Juergen Rauschenbach DFN Germany
Victor Reijs SURFnet The Netherlands
1. APPROVAL OF MINUTES
The minutes of the TF-TANT meeting held on the 29th and 30th of
March 1999 were approved.
2. STATUS OF QUANTUM & TEN-155
Howard reported the connection to Portugal was currently being
tested and delivery was planned for the 24th of June. There were
however, some problems with the connections to Ireland and
Luxembourg. The line to Ireland had been installed, but the ATM
switch would not be installed for another four weeks. A temporary
solution was being planned in the meantime. There were also
equipment problems in Luxembourg which meant that connection had
been delayed.
A contract had still not been signed with the Slovenian PTO.
Negotiations had broken down and alternative suppliers were being
approached. Whilst there was a telecommunications monopoly in
Slovenia, this did not extend to data provision.
The TEN-155 Managed Bandwidth Service (MBS) was currently in the
beta-test phase which was scheduled until the end of June. The SUSIE
(Germany/Switzerland), EDISON (France/Germany), METODIS
(France/Germany), ENCart (Austria/Hungary) and TF-TANT (MPLS
experiment) projects were piloting this. This had raised a number of
issues about the management chain, which involved up to three
separate organisations (DANTE, local NRNs and local institutions).
It was clear that unified management procedures would be required in
future.
As a result of the upheavals in the European Commission, it was
likely that the Fifth Framework Call for Proposals would be
postponed from September 1999 until January 2000. The consequence of
this may be problems with continuity of funding between QUANTUM and
the next generation network.
The QUANTUM Policy Committee (QPC) had met on the 1st of June and
had expressed a lot of interest in the QUANTUM Test Programme (QTP).
They had decided that QTP deliverables should be peer reviewed
before being published in order to validate the quality of the
results. In addition, they had asked whether a joint workshop could
be held with TF-TANT in October, so they could evaluate progress.
Wilfried said he did not have a problem with peer reviews
themselves, but he wanted reviewers that would provide the group
with constructive guidance. During the TEN-34 project, criticisms
were only voiced after the test programme had been completed.
Kevin asked whether the QPC would find the peer reviewers, or
whether the TF-TANT group was expected to do this. Howard believed
this was the responsibility of the QPC.
3. FORMAT OF FUTURE MEETINGS
Christoph said the group had agreed to have two parallel sessions
during the current meeting. There were too many experiments for them
all to be discussed during a plenary session, and the parallel
sessions would allow more time for thrashing out technical details.
The group however, still needed to decide the optimum format of
future meetings. One option was to have separate experiment meetings
on one day, with a plenary meeting on the second day. Alternatively,
separate experiment meetings could be held in a common location, or
even in different physical locations.
Tiziana proposed that a general meeting could still be held, but
individual experiment meetings could arranged as necessary. Wilfried
however, said there would then be very little incentive for
participants to come to the general meeting. In the long-term, this
might affect the work programme as the current format provided
opportunities for receiving feedback and finding participants.
It was agreed that future meetings would have one day for separate
experiment discussions, followed by a joint meeting the next day for
inter-experiment coordination and progress reporting. This would
hopefully strike a good balance between the need to discuss
technical details, and allowing participants to gain an overview of
all the experiments.
4. EXPERIMENT DISCUSSIONS (Parallel Sessions)
4.1 QoS Monitoring
Philip discussed the different approaches to monitoring QoS. This
could be conducted actively by inserting test traffic into the
network (e.g. Surveyor equipment), or passively using techniques
such as NetFlow. The parameters that should be measured included
one-way packet loss, one-way delay variation, link utilisation, link
bandwidth, EF commitments and EF reservation load. It was also
suggested that interface discards, one-way packet delay and burst
throughput should be measured. The goal was to define a meaningful
measurement architecture that produced consistent traffic reports.
Tiziana asked what techniques could be used to conduct these
measurements. Philip replied it might be possible to utilise the
Surveyor or RIPE test traffic equipment. Another alternative was to
use NTP, although this was not very precise.
Tiziana went on to ask whether the Surveyor equipment could generate
TCP streams. Philip thought this was possible, but they not yet
managed to get their GPS antenna to work. It was also possible to
shape traffic although the equipment was not intended to be a
traffic generator.
It was thought the Surveyor equipment would be more suitable for
this experiment because the RIPE Test Traffic Project was aimed at
commercial networks where confidentiality was important. Kevin said
he had some contacts at Advanced Networks (who developed Surveyor)
and would ask about the possibilities of loaning some equipment.
ACTION 4.1 - Kevin Meynell
Jean-Marc said it might be possible to loan more Smartbits equipment
from Netcom. They had been quite helpful during the MPLS experiment,
although they had just started a new activity. He would approach
them about this.
ACTION 4.2 - Jean-Marc Uze
Roberto mentioned the University of Cambridge also had a measurement
project. They had some prototype equipment they would like to trial.
It was agreed that a permanent DiffServ network should be
established after the initial tests. This should not be too
complicated and should not involve more than about five sites. A
topology had already been specified in the experiment proposal.
Tiziana mentioned she was looking for someone to help her coordinate
this experiment. Roberto volunteered for this.
4.2 IP Version 6 (notes from Christoph Graf)
Simon N reported that Telebit had shipped one of their routers to
the Amsterdam PoP, which was currently awaiting installation by
DANTE (http://www.tbit.dk/quantum/slides/tftantbologna/index.htm). A
pTLA had been obtained (3ffe:8030:/28, QTPVSIX) and plans
for the MBS overlay were being finalised. They would start
connecting participants shortly. There were also plans to connect to
the 6REN in the future.
Simon N asked whether end-to-end tests should be conducted, and
whether additional services should be tested in a coordinated
manner. The group was in favour.
Wilfried proposed the experiment also cover address allocation,
peering and deployment of IPv6 in existing infrastructures,
tunnelling, DNS and reverse DNS.
It was agreed the first phase of the experiment should connect
Austria and Switzerland to the 6Bone via TEN-155 during the first
phase. 512 Kbps to 1 Mbps bandwidth would be adequate, but this
should be on a permanent basis if possible. The remaining
participants would be connected during the second phase of the
experiment.
The following address usage was agreed: pTLA = 28 bits, NLA = 6
bits, SLA = 14 bits and subSLA = 16 bits. An AUP was not considered
necessary as this was only a pilot service.
Simon N asked each participant to supply information about their
available equipment, bandwidth and manpower. DANTE was also asked to
create a mailing list for this activity.
ACTION 4.3 - All IPv6 experiment participants
ACTION 4.4 - DANTE
4.3 Differentiated Services
Christina gave a presentation on the MONARC project. This was a
distributed computing application for processing high-energy physics
data, and could potentially benefit from DiffServ.
Silvia asked whether the MONARC application was DiffServ enabled.
Tiziana replied it was not, but DiffServ could be enabled on the
routers.
Tiziana thought testing an application that was only used by a
specialist user community would not be of much interest to most
people. Leon however, said this was an extremely important project
and worth considering.
Mark Jansen gave a presentation on the DiffServ tests conducted at
the University of Utrecht (http://www.phys.uu.nl/~jansen/
diffserv.ppt). Their results showed that bi-directional TCP streams
from multiple source to multiple destination provided the best
performance.
Werner Almesberger gave a presentation on the Linux DiffServ
implementation. The latest version (http://icawww1.epfl.ch/
linux-diffserv/dist.html) had a modular framework and supported
traffic control, classification, scheduling and policing. Policy
management, ingress policing and improved classification was still
being worked on.
Tiziana asked about the scripts necessary to use Linux as a DiffServ
router. Werner replied the scripting language was a bit messy at the
moment, but help could be obtained from the mailing list (see
http://lrcwww.epfl.ch/linux-diffserv/ for details).
Tiziana then presented her proposal for the differentiated services
experiment. The first phase was scheduled from the 21st of June to
the 28th of August. There would be a single domain consisting of
INFN, Stuttgart and Utrecht running Cisco and Cabletron
implementations of DiffServ. Basic QoS measurements using Smartbits
equipment would conducted during this phase.
The second phase would involve multiple domains running different
DiffServ implementations (IBM and Linux). This would investigate SLS
definitions, LDAP policy servers, traffic shaping and policing,
remarking and heterogeneous platforms. The third phase would test
real applications over DiffServ (e.g. MONARC and video-streaming).
The dates for these phases however, still had to be decided.
The requirements for each participant were a Smartbits box and a
test workstation with NetPerf, mgen, ttcp and NetXpert installed. It
was also necessary for configure NTP and a common account for SSH
access. At least two sites would require a GPS antenna for the
Smartbits boxes.
Silvia thought the proposal was a bit ambitious. Interoperability
testing was very time consuming and needed a lot of support from the
vendors. She suggested the number of domains should be reduced.
Simon L commented that SWITCH would not have a native ATM service
until the end of August.
Philip mentioned that Torrent had just been acquired by Ericsson and
were keen to participate in the experiment. Their DiffServ
implementation was due in October.
Werner asked why addresses were allocated in blocks of ten. Tiziana
replied this was for aggregation purposes. Simon L thought blocks of
eight would be better in this case, especially if static routes were
used.
Paolo said he did not want to use reserved addresses because he had
researchers who wished to access the network. The network should be
routeable even if it was still physically isolated.
Werner suggested the ISABEL video-conferencing application developed
by the University of Madrid may be an interesting application to
test over a DiffServ network. It had been ported to Linux and would
be useful to most people.
Tiziana asked about participants unavailability during the first
phase of the experiment:
CERN Available all weeks
DANTE Available all weeks
INFN Weeks 29-30, 33-34
RedIRIS Weeks 33-34
Uni. Stuttgart Week 34
SWITCH Week 29, 31
Uni. Twente Week 30, 31
Uni. Utrecht Available all weeks
Tiziana said further activities might include the introduction of
other vendors (e.g. Nortel, Telebit and Torrent) into the DiffServ
cloud, interworking trials between QoS-capable MPLS clouds and
DiffServ clouds; testing of bandwidth brokers; and interconnection
with other testbeds such as the QBone.
4.4 Multicasting (notes from Christoph Graf)
Jan gave a presentation on the proposed TEN-155 multicast service
(http://www.dante.net/mbone/mcast99/mphase2.html). The PIM Sparse
Mode cloud would contain nodes in the Netherlands, Denmark and
Sweden. NRNs were being encouraged to migrate to MBGP as parallel
DVMRP caused forwarding loops which could not be maintained in the
longer term.
There were some interoperability problems between Cisco IOS 12.0 and
11.1.25cc. Participants were therefore requested to run IOS 12.0
(Wilfried thought 12.0.48 was okay as well). Other problems were
that NRNs were not supplying much information to generate MBGP
routes, DVMRP interaction did not work well, and multicast running
over an ATM full mesh did not make efficient use of the underlying
ATM infrastructure.
The group discussed the last problem, but were not very happy with
any of the current solutions. Nevertheless, a proposal by Steffen
Baur looked promising and would be tested.
[Note: the subsequent results of this test can be found at:
http://www.dante.org.uk/mbone/mcast99/mphase2.html]
5. REVIEW OF EXPERIMENTS
5.1 ATM Signalling
Jan reported the service description was now available on the WWW
(http://www.dante.net/nep/dante-kpn/docs/1999/DK-99-21.html). KPN
had agreed to most requirements, and there was a lot of freedom to
specify parameters. One problem though, was a lack of support for
SBR1. This meant that bandwidth usage could not be restricted unless
CBR was used.
The proposal for the acceptance tests was also available on the WWW
(http://www.dante.net/nep/dante-kpn/docs/1999/DK-99-16.html). Some
dates still needed to be specified, and the acceptance criteria
still had to be agreed with KPN.
DANTE required some feedback from the experiment participants. Did
they plan to use production interfaces on their switches; did they
have a workstation with an ATM interface that supported LANE; and
did they plan to use their own addressing scheme?
Jean-Marc and Guenther said they planned to use their own addressing
schemes. Jan was also asked to include the KPN addressing plan
(NSAP/AESA E.164) in the experiment proposal.
ACTION 4.5 - Jan Novak
5.2 Policy Control
Leon reported the research phase of this activity was about to
finish. The next stage was to develop plans for testing equipment
using LDAP and policy management applications. A number of vendors
had LDAP-compliant products, including Cabletron, Cisco, IBM, Nortel
Networks, Telebit and Torrent. Any suggestions of other equipment
that could be tested were welcomed.
5.3 Multicasting (IP and ATM)
Robert reported the TEN-155 multicast service was being migrated to
MBGP. They hoped to move to a fully operational service in time for
the Oslo IETF.
The ATM Point-to-Multipoint tests would be scheduled during the
TEN-155 acceptance tests in September. The participants would be
DFN/RUS, ACOnet, CESNET, INFN and RedIRIS. An experiment description
was currently being drafted, and this could be discussed at the next
meeting. The requirements at each site were a Cisco router with an
ATM interface (AIP or ATM-PA3), an ATM switch terminating a VP to
TEN-155, and a test workstation.
5.4 IP over ATM
Roberto said there was little to report since the last meeting. KPN
had been informed by Ascend that a fix was now available for SBR/SCR
and Y2K problems in their switches. Implementation of this fix would
therefore needed to be coordinated by the NOC.
5.5 Flow-based Monitoring
Simon L reported that DANTE had opened a test account on a
workstation located at the TEN-155 PoP in Switzerland. This received
a copy of the NetFlow accounting data from the router also located
there. The software installed to analyse the data included cflowd,
Cisco FlowCollector/FlowAnalyser and Fluxoscope (A NetFlow
accounting tool developed by SWITCH).
The aim was to investigate how accurate traffic statistics could be
produced for exchange points. This could help facilitate
volume-based charging (such as the JANET scheme), detection of
network abuse (e.g. smurf attacks), identify long-term trends (e.g.
application mix, new applications, interesting source/destination
networks), and detection of routing anomalies. The available tools
would be described, compared and deployed where appropriate.
Simon N asked why multicast was not being used to collect the data.
Simon L replied they had been unable to get it to work.
5.6 Route Monitoring
Simon L reported that IRRd (routing registry mirror server), mrt
(multi-threaded routing toolkit) and the RAtools software had been
installed on a workstation located at the TEN-155 PoP in
Switzerland. This had been difficult in some cases where the
programs had required root privileges. The next stage would be to
set-up routing registry mirror servers so the actual TEN-155 BGP
routing table could be compared with these.
5.7 IP Version 6
Simon N summarised the outcome of the earlier IPv6 session
(http://www.tbit.dk/quantum/slides/tftantbologna-plenum/index.htm).
The aim was to conduct end-to-end tests over 0.5 Mbps VCs, and to
run some native services such as DNS. These activities would
continue until the end of the QUANTUM project. It was hoped that
other router vendors would also participate.
Bert mentioned that IBM had routers available for loan that were
IPv6-compliant. CERN were responsible for distributing these.
Jean-Marc asked whether RENATER could connect using their own IPv6
addressing scheme. Simon N replied this would not be a problem.
5.8 RSVP to ATM Mapping
Tiziana said this activity would commence when ATM signalling was
available on TEN-155.
5.9 MPLS
Jean-Marc gave a presentation on the MPLS trials conducted during
the previous three months (http://www.renater.fr/jmu/QTP/MPLS-tf-
tant-180699/). Configuration was straightforward and the stability
of MPLS had improved since the previous tests. Performance had been
tested with low bandwidth links using Netcom Smartbits 2000, ttcp,
mgen, NetPerf and ping, but had consistently proved to be
unacceptable. This required further investigation.
The next stage was to conduct more performance tests, and
investigate PVCC configuration, Tag QoS, and management issues. This
would probably happen in the Autumn.
Jean-Marc thanked everyone who had participated in this experiment,
particularly Victor Reijs.
6. TF-TANT OVERLAY NETWORK
Roberto said the previous meeting had postponed the decision on
whether a permanent overlay network should be established for the
TF-TANT experiments. There was a discussion of the problems
experienced during the MPLS experiment, and whether a permanent
overlay network would help future experiments. A possible topology
was also presented.
Wilfried pointed out that logistical problems had recently cost them
a lot of time. A switch had been misconfigured to send and receive
cells on different interfaces. This may be due to the fact that
different switch vendors had their own interface numbering schemes.
It would also be useful to have a common agreement for loopback
tests at management boundaries.
Robert asked about the response time for setting-up VPs on the
Managed Bandwidth Service. Howard replied they aimed to fulfil
requests within two hours. They did not however, have any control
over the response times of local networks.
Wilfried said he would require more information about the MBS and
the duration of the TF-TANT programme in order to put management
procedures in place in Austria. Christoph suggested that questions
be sent to the MBS Operational Mailing List.
It was concluded that the establishment of a permanent overlay
network would not really help. Many problems were local to the NRNs,
permanent connections sometimes incurred costs, and only one ATM
traffic class would be available.
7. DATE OF NEXT MEETING
The next meeting was provisionally arranged for the 30th of
September and the 1st of October to coincide with the next QPC
meeting. The first day would be reserved for specific experiment
discussions, whilst the second day would be a joint session with the
QPC. The venue would probably be Amsterdam.
Kevin said that TERENA would be unable to host this event as their
meeting room could only hold a maximum of twenty-five people.
Schiphol Airport was suggested as an alternative.
8. ANY OTHER BUSINESS
Christoph thanked Tiziana and INFN-CNAF for hosting the meeting.
9. ACTIONS FROM LAST MEETING
3.1 Christoph Graf to circulate EU funding notice/disclaimer on the
mailing list.
- Done.
3.2 Christoph Graf to circulate a questionnaire about MBS
requirements on the mailing list.
- Superseded.
3.3 Tiziana Ferrari to incorporate bandwidth brokers into the
Differentiated Services experiment proposal.
- Ongoing, but nearly completed.
3.4 Roberto Sabatino to describe laboratory tests in the IP over
ATM experiment proposal.
- Closed. Unable to publicly disclose these.
3.5 Victor Reijs to send URL of document discussing the use
DiffServ in conjunction with RSVP to the mailing list.
- Done.
3.6 Roberto Sabatino to obtain approval for the TEN-155 multicast
network from the QUANTUM Policy Committee.
- Done.
3.7 Guenther Schmittner to ask KPN whether they will support PNNI.
- Done. KPN would not be supporting this.
3.8 Victor Reijs to draft document expressing the concerns of the
research community about STM-4c.
- Ongoing.
3.9 All MPLS experiment participants to arrange ATM connectivity to
TEN-155, complete MBS questionnaire, put diagram of local Tag
architecture on the Web, specify delivery address for loan
equipment, and sign Non-Disclosure Agreement.
- Done.
3.10 Jean-Marc Uze to propose backbone infrastructure and VPN
set-up, circulate the configuration document, and organise
Smartbits tutorial.
- Done.
3.11 Cisco to check the hardware configurations, determine software
availability, and provide feedback on test plan.
- Done.
2.1 Simon Leinen to contact Tiziana Ferrari to determine what
activities could be included in the RSVP to ATM SVC Mapping
Experiment.
- Done. No conclusion was reached.
2.2 Robert Stoy and Jan Novak to specify the TEN-155 multicast
facilities.
- Done.
2.12 Experiment Leaders to specify equipment requirements in their
proposals as soon as possible.
- Done.
OPEN ACTIONS
4.1 Kevin Meynell to contact Advanced Networks about the
possibility of loaning some Surveyor equipment.
4.2 Jean-Marc Uze to approach Netcom about the possibility of
loaning more Smartbits equipment.
4.3 All IPv6 experiment participants to supply information about
their available equipment, bandwidth and manpower.
4.4 DANTE to create a mailing list for the IPv6 experiment.
4.5 Jan Novak to include the KPN addressing plan in the ATM
signalling experiment proposal.
3.3 Tiziana Ferrari to incorporate bandwidth brokers into the
Differentiated Services experiment proposal.
3.8 Victor Reijs to draft document expressing the concerns of the
research community about STM-4c.
Contact: nep@dante.org.uk