Minutes of the 5th TF-NGN meeting

15-16 October 2001, National Technical University of Athens, Greece

Valentino Cavalli, Issue 1
 

Table of Contents

Attendees

Name Organisation Country
Wim Barbaix  Alcatel Belgium
Kurt Bauer University of Vienna Austria
Alain Bidaud Crihan/Renater  France
Artur Binczewski  PSNC/POL-34  Poland
Mauro Campanella GARR-INFN Italy
Graça Carvalho Cisco Systems Portugal
Valentino Cavalli (Secr) TERENA -
Tim Chown Univ. of Southampton United Kingdom
Pascal Drabik European Commission  -
Larry Dunn  Cisco Systems  USA
Tiziana Ferrari INFN-CNAF Bologna Italy
Avgust Jauk  ARNES  Slovenia
Joop Joosten  CERN  Switzerland
Dimitrios Kalogeras GRnet Greece
Olav Kvittem  Uninett  Norway
Haris Laskaridis  GRnet  Greece
Simon Leinen SWITCH  Switzerland
Ladislav Lhotka CESNET Czech Republic
Athanassios Liakopoulos  Grnet  Greece
Orla McGann  HEAnet  Ireland
Octavio Medina  IRISA  France
Kevin Meynell  TERENA  -
Marcin Michalak  STC-ULB  Belgium
János Mohácsi  DANTE -
Ramin Najmabadikia  BELNET  Belgium
Havaliotos Posos  CTI  Greece
Jürgen Rauschenbach DFN-Verein Germany
Victor Reijs SURFnet & HEAnet  The Netherlands & Ireland
Rudolf Roth GMD FOKUS Germany
Roberto Sabatino (chair) DANTE -
Rina Samani  UKERNA  United Kingdom
Pavel Satrapa CESNET Czech Republic
Christian Schild   JOIN-Univ. of Münster  Germany
Mark Schäfer DeTe-Systems Germany
Afrodite Sevasti  GRnet  Greece
Nicolas Simar  DANTE  -
Wim Sjouw Univ. of Utrecht The Netherlands
Trond Skjesol Uninett  Norway
Miguel Angel Sotos RedIRIS  Spain
Robert Stoy  DFN  Germany
Szymon Trocha  PSNC  Poland
Bernard Tuy  RENATER  France
Chrysostomos Tzioyvaras  GRnet  Greece
Jean-Marc Uzé  Juniper Networks France
Stig Venaas Uninett  Norway
Franz Widhofner  University Linz/Aconet  Austria
Steven Williams  UKERNA/UW Swansea  United Kingdom

Apologies

Name Organisation Country
Yves Schaaf  RESTENA  Luxembourg
Wilfried Woeber  ACOnet  Austria

Online presentations: http://www.terena.nl/task-forces/tf-ngn/presentations5.html
 

1. Introduction and GÉANT update

Vasilius Maglaris from GRnet welcomed the meeting attendees by addressing significant networking innovations in Greece. He said that the National Technical University of Athens, NTUA, has four GigE switches in the campus and about fifty Ethernet switches. They have deployed 300 Km of DTP garment and10 km of fibers. GRnet has six POPs in Greece and more will be operational in the future. In conjunction with GÉANT, two carriers provide 155 Mbps that will be upgraded to 622 in 2002. GRnet will upgrade the national backbone to IP over WDM and will create a MAN in Athens based on three mega routers, it is not yet know from which vendors.  Fifteen POPs will have Lambdas. The fibers are already installed and GRnet are just waiting for the equipment. Vasilis mentioned a bid aiming at deploying Lambdas all over Greece in order to replace the current STM-4 links. GRnet was currently negotiating with one operator only, OTE, the Greek PTT.

Roberto Sabatino from DANTE reviewed the list of open actions. Three actions from previous TF-NGN meetings were still ongoing but have been superseded by new ones. Action 1.13 was on Robert, Tiziana and Lada to refine the workplan for QoS & Multicast. Robert made a presentation at the meeting in Athens and estimated to work out a plan within three months.

- Action 5.1, Robert Stoy to provide detailed work plan for QoS & Multicast before the 6th TF-NGN meeting.

Action 1.18 was on Tiziana. Mauro, Roberto and Hervé to discuss about the loans of  equipment. The action was substantially done, but Tiziana said she was interested in investigating about options from Juniper Networks and asked to keep the discussion open.

- Action 5.2, Tiziana to coordinate discussion about loan of Equipment.

Action 2.1 was on the MPLS activity group to check specific features of MPLS traffic in connection with BE traffic. The action was not done, but Alain Bidaud said that the whole activity was dormant mainly due to lack of manpower at Crihan. Roberto observed that the whole TF-NGN group should discuss how to tackle this problem.

- Action 5.3 the TF-NGN group to discuss how to carry over the MPLS activity.

The full list of action is given at the end of this report. In the following only those with specific comments are mentioned. Action 4.9 was on Wim Barbaix to collect proposals for interconnections and experiments with the ATRIUM project. Wim said he had received one proposal about QoS testing, and was still willing to encourage people to submit new ones. Action 4.10 was related to functionality of the Cisco Engine Cards. Graça said she had received some requests and recommended people individually to set up the equipment with the latest cards. Graça added that the equipment Cisco is going to provide on loan to TF-NGN was ready to be shipped, however Cisco was still waiting for determined plans about condition and duration of the experiments.  Action 4.13 regarded requirements for experiments to be carried out on the Garr-G test bed. Mauro said he did not receive any request however the test bed was still available to TF-NGN members until November 2001. Action 4.14 was on Jean-Marc to provide information about support for shaping in Juniper routers. At the meeting he said that no shaping feature is supported in the software at the moment and that he will update the group when it will become available. The last discussion was related to Action 4.15 on Tiziana to evaluate the Pioneer test bed for shaping experiments. She informed the group that the decision was to test other equipment and had asked Juniper and Cisco to use the higher-end equipment. She added she was hoping to get it as soon as possible to start the tests.

Roberto updated the audience about the planned GÉANT topology and said that some countries including France, Italy, Germany and Belgium had already successfully past acceptance testing, while some others were currently being tested. Finally there was still a small group of countries that was not yet started. There had been very few issues with circuit delivery. Most problems were related to POP setup. DANTE was working on the European Distributed Access - interconnection with other NRENs - in London and Frankfurt to the GÉANT co-location in New York at 60 Hudson St. where GÉANT would peer with Abilene, ESNET and CANARIE.

In order to minimize the use of GÉANT for commodity Internet DANTE was also running a tender for commodity IP access in Europe. Roberto said that Germany, Spain and Italy will bypass GÉANT and will have connectivity delivered in the country, whereas other countries will use GÉANT as transit.  The ATM circuits will go away on November 30 this year, therefore DANTE are working hard in order to ensure that the GÉANT service becomes operational before that date.
 

2. Updated Technology Roadmap

Roberto Sabatino reviewed an updated version of the GÉANT technology roadmap for 2002 that was discussed at the TF-NGN meeting in Tromsoe. The roadmap has to be submitted as a Deliverable to the EC later this year. He clarified that the plan was not yet finalized and it was still possible to add new topics. Jean-Marc replied saying that one interesting item to be investigated might be the QBone Scavenger Service.

The main discussion points were raised by Tiziana, Dimitrios, Simon and Artur. Tiziana asked for a clarification about the premium IP service test bed. Roberto said that a test for the streaming community would be run for the first time in November 2001 over an international production network. The pilot would run until May 2002. Dimitrios suggested that MPLS TE could be provided as a service where a static LSP is provided to end customers. Roberto pointed out some confusion arising from the fact that MPLS TE is listed in the roadmap along with other services, but MPLS TE is not a service, it is rather going to be used to implement other services. Simon suggested of listing it in a different way, so that instead of MPLS TE, VPNs or Virtual Leased Lines are mentioned as a service. After such a clarification Dimitrios rephrased his suggestion by saying that it would be interesting to deploy VPNs without leaving the end customers to do the same. Artur suggested that interoperability tests should be added to all topics in the roadmap. Roberto remarked that there is already an item on testing new equipment in the roadmap. In reply, Graça pointed out that interoperability of equipment regards mainly protocol interoperability and she agreed with Artur that there should be some specific investigation on each service.
 

3. Multicast

Ladislav Lhotka from CESNET briefed the group about recent developments on the improvement of multicast services. He mentioned that the Deliverable 9.2 is available on the DANTE web page and invited people to comment. He recalled the graphic presentations of multicast beacon time series developed in parallel by Roman Lapacz from the Poznan Computer Centre and Rober Stoy from DFN. He said that the Beacon agent written in C is now operational under linux http://zma.ten.cz:8378. However, it still needs to be further tested and ported to other architectures. More things to be done in the near future include deployment of IPv4 multicast in GÉANT and the need to discuss it in relation to the overall network architecture. Application level authentication for agent communication needs to be added. It was suggested that Beacon agents are installed not only at GÉANT POPs but also at the NRENs access points, so that Beacon can monitor traffic on these links, using a separate group that that used by GÉANT.
 

Robert Stoy from DFN presented some experience of QoS with IP multicast. He wrote extensions to the Beacon measurement environment in order to analyse and visualise statistic measurement data. Visualisation of statistics on the TF-NGN Beacon session are provided at http://jaspis.noc.dfn.de/tf-ngn. Within DFN a similar Beacon session with statistics visualisation is running. This DFN-Beacon session is used to measure and visualize Multicast QoS on the DFN Multicast service. The measurement environment comprises workstations running a beacon client, connected to the core routers within each core network node. Also some selected customer's workstations running a beacon client are participating for measuring end-to-end Multicast QoS. Robert said more interested customers can still join the infrastructure. The experiments involved among others Multicast BE traffic forwarding and Multicast Priority forwarding. In the current phase work is done mainly on the improvement of the BE multicast traffic forwarding. Multicast Traffic Process Switched has high packet loss and was not recommended by Cisco. Robert concluded by saying that the measurement environment is in place and the current challenge is to improve BE multicast in order to provide small packet loss on the GSR, OC12 GigE infrastructure.
 

Artur Binczewski from Poznan presented results of SSM tests carried out at the Poznan Computer Centre in Poland. Incidentally he mentioned that he heard from Cisco that the new version of packet over SONET for OC-48 had been optimized for multicast traffic. The goal of the Polish experiments was to run SSM on native multicast and test it with multimedia stream protocols. URD Host Signaling tests used Beacon software for several purposes. The Beacon software is very good and stable, and can be used effectively to monitor traffic between clients. For application level tests one should specify route and sources were traffic is generated. They can implement SSL protocols. Artur mentioned that at Poznan they continued adding features to the Beacon software. They had provided the history feature to the original creators and other developers, but they did not receive any answer yet. In addition he mentioned the development of statistic features. Future work on Beacon development would include mtrace integration and some other new projects like provision of an open source plug-in for browsers and multicast support, and finally unicast-multicast translation applications. Simon remarked that the Beacon software is reachable over IPv6. He also added that while he was interested in the SSM test he still had some doubts about the mtrace integration and suggested to use other protocols like IGMPv3.
 

4. Flow based Measurement

Simon Leinen from SWITCH updated the audience about the latest developments since the meeting in Tromsoe. He said that input to the deliverable 9.4 had been provided and the deliverable was complete. It contains a chapter on flow-based monitoring containing sections on measuring peer-traffic, new mechanisms for BGP based accounting and the identification of problems experienced with Netflow.

On the standardization side, Simon reported that the Ioflow Export Information working group has changed its name from ipfx to ipfix and is now in the operation and management area. At the London IETF there was a BOF with presentation of ongoing work, including netflow-v9b. Simon has made an attempt to update the FloMa pages with new references but he said it was not easy to find them. He added some details about technical tests carried out in Switzerland. Access to the workstation in the mean time has been closed down for security reason, but he said there is no reason not to enable access in the future, and it would be made available again to the monitoring group depending on their actual activity requirements. Besides FloMa, Simon mentioned the ongoing SNMP standardisation effort, notably documented in ietf-diffeserv-mib-14. The activity is reaching consensus in the working group that is also busy in revising BGPv4 MIBs.
 

5. AF based testing

Octavio Medina from IRISA presented his study about TCP behaviour with Assured Forwarding. The goals of his tests were basically to check bandwidth sharing between aggregates, to determine how to protect TCP from UDP flows and to calibrate the parameters of three-colour marker (TCM).

The tests were carried out on a network topology where Solaris workstations were used to generate the flows going to Cisco 7500 routers, there was no congestion at this point, bottleneck link had WRED. All test results obtained were based on homogeneous RTT. Background UDP flows and forty TCP flows were injected into four aggregates. Four markers were used in the input interface. Four different congestion levels were tested, bringing the experiment very close to the maximum link capacity. The right WRED configuration parameters were determined in order to ensure that AF works correctly. TCM configuration used two token buckets.

A number of tests were performed. Test T0, with no AF at all, showed that bandwidth is shared equally among aggregates but TCP is considerably affected by UDP and there is a need to control the distribution of bandwidth. Test T1 defined initially the size of the first token bucket (alpha) and showed that results were almost independent of the size. The behavior of the second token bucket (beta) was then defined and the experiment was carried out with different congestion levels. Octavio found that if beta is too small then there is a large amount of red packets, whereas if beta is too big then there are too many yellow packets. The result is that differentiation is reduced by one level and there is not enough bandwidth to transport yellow packets. Test T2 was done by increasing the size of the second token bucket resulting in increased bandwidth allocated to TCP. Test T3 was repeated with selected values of alpha (0.1, the smallest one) and beta (2.5), in this case it was clear that differentiation is more effective when UDP is active. Under congestion there is good differentiation but the aggregate fails to obtain the desired rate. In summary, it was clear that AF is only worthwhile in distributing packet-loss in case of congestion. With no congestion AF is not needed. Other conclusions were that TCP can be effectively protected from UDP flows if different marking is used and bandwidth-sharing can be controlled at the aggregate level, especially on congested links. Octavio said he plans to continue doing more tests in the future but will try other marking schemes, and produce a number of scenarios with different RTTs.
 

6. IPv6

Tim Chown from the University of Southampton introduced the IPv6 session by providing a quick overview of the IPv6 test network, a star topology with a telebit TBC2000, 17 native IPv6 connections over ATM PVCs and 10 tunneled. The network was not operational at the time of the meeting pending upgrade. New connections were due to HEAnet and to Poznan, some other minor issues regarding routing were also affecting it, like route flapping between CERN, RedIRIS and SWITCH. Tim continued the introduction by mentioning a number of initiatives around IPv6 in Europe, including the 6NET project, a meeting between European organizations and the WIDE project that took place after the August IETF in London, the NGN-LAB project. The agenda would also include an update on DSTM, firewalls, other partner updates, the status of the M5 router provided by Juniper Networks, the gtpv6 network topology. He also said that the GÉANT deliverable regarding IPv6 was submitted to the EC in August 2001, that an addendum for some areas was being prepared and supposed to be submitted in November. A draft version of the addendum had been mailed to the gtpv6 mailing list recently.
 

Graça Carvalho from Cisco Systems updated the audience about the status of 6NET, a three-year EC project with many participants from TF-NGN, including DANTE and TERENA. The contract negotiation ended three weeks before the TF-NGN meeting, and the contract was going to be sent to the partners. It may be starting in December 2001 or January next year depending on how fast the signature process will be. Pascal Drabik from the European Commission said that the contract will start the month after the Commission signature and that they would try to speed up the process internally with the target of starting on the first of December. Cisco Systems coordinates the project, which has more than thirty participants. The negotiation process was longer than planned mainly because of the additional paperwork needed when changing many university subcontractors into formal contractor. However, as a result of the process, the description of the work is detailed in much better way in the final technical annex as it was in the proposal. The project is worth a contribution from the EC amounting to 17Million EUR, making 6NET the biggest IST project after GÉANT, therefore it is quite natural that the Commission requested the consortium to detail the work very specifically. Graça said that the Commission services were currently busy in checking contractual details against their internal information and if they found discrepancy they would contact the participants that should be ready to provide any requested clarification in a very short time, in order not to delay the start of the project.

The 6NET project is articulated in seven work packages. The first one is about the infrastructure, other work packages are more technology oriented and finally there are some management and dissemination work packages. The goal is to provide native IPv6 connectivity from day 1, to this aim DANTE will approach the GÉANT suppliers in order to use the same providers and the same contracts as GÉANT in the 6NET infrastructure, but this will be done only after the GÉANT implementation. In some cases, like in Greece, tunneled IPv6 seems to be unavoidable. Eight links at 155 Mbps will be available for three years starting from month 3 (likely March 2002). Four 2.5 Gbps native circuits on GÉANT will be available on month 13, however precise end points still need to be discussed and at that time encapsulated circuits may be upgraded. What will be available in each country is still to be discussed and depends on what will be available and the particular capacity in those countries. There is also an option for Greece to have lower capacity but on a native link.
 

Tim Chown reported that there was a meeting between EU and Wide in August at the University College London about collaboration on IPv6 between Europe and Japan. He said that in November he is going to travel to Japan with Peter Kirstein from UCL in order to establish some collaboration with KAME and about multicast IPv6 in Japan. Bernard Tuy was the organizer of the London meeting and reported that the conclusion of the meeting was to have one such a meeting organized after every IETF. Bernard said his suggestion was for a half-day duration and added that anybody in the group who is interested to attend it should get in touch with either him or Tim.
 

Marcin Michalak is working at the Universite Libre de Bruxelles, ULB-STC, on the NGN Lab project. He recalled that the project is meant to provide connection between the EuroDemo laboratory in Brussels and the Mclab in Basel. The infrastructure is based on ATM links connecting Belnet, DANTE, SWITCH and the University of Basel. The main idea is to exploit native IPv6 connection to test international links as a common test bed for different IPv6 applications. Marcin said everybody is welcome to join by contacting him, Rosette Vandenbroucke or Professor Paul van Binst at ULB. He also handed over copies of the NGN-LAB press release. The main issue is a time constrain because the test bed is running on the TEN-155 ATM backbone that is going to disappear before the end of the year, so anybody willing to use it must do it immediately. On the other hand it is now critical to start thinking how to continue the provision of the NGN-Lab test bed on a different infrastructure. Some hypotheses were made, including tunneling, but it was needed to understand what SWITCH can do internally first. Roberto Simon and Marcin agreed to investigate into the continuation of NGN Lab.
 

Octavio Medina explained that the Dual Stack Transition Mechanism, DSTM, is meant to ensure communication between IPv4 and IPv6 in such a way that IPv4 only applications can run even on a IPv6 only network. Both IP stacks are needed on the host, but IPv4 is configured only when is needed by the applications and all IPv4 traffic is encapsulated on DSTM. In DSTM there are two ways of managing address allocation: manual configuration and dynamic allocation. It is a reliable mechanisms, but has some side effect depending on the roaming scenario, for instance when it occurs that there are no IPv4 addresses left. Octavio compared DSTM with NAT-PT and said the former has some advantages with respect to the latter, which has the same problems as classic NAT. DSTM gateway and server have been developed under BSD at INRIA in France. Clients are available on other platforms too.
 

János Mohácsi from DANTE is in charge of investigating about firewalls in the TF-NGN IPv6 working group. He remarked that security needs to be significantly improved with respect to the current situation. Due to time constrains he had to shorten his presentation about the IPv6 architecture and firewalls and limited his speech to some essential messages, like an urge to carefully designing firewall architectures in IPv6 environments. NAT would not be needed. János proposed not to filter local addresses but only global ones. Also, firewalls should be prepared to distinguish (to follow the chain of) v6 headers.

János briefly reported some results about his evaluation of available firewalls. In general the architecture of IPv6 firewalls is not very modern, there is no proxy support, and auto-configuration is not well supported, but support IPv6 extension headers. The main problem with Ipfilter is that it does not provide support for IPv6 extension headers and does not support ftp proxy support, however it has a quite complete architecture. Netfilter has a good extensible architecture, proxy support is available only via external kernel programming, but support for icmpv6 is rather weak. Cisco Access List has good commercial support, but only addresses filtering. János tested interoperability of filtered application on other firewalls, like 6wind and ip6fwtk. There was a question about testing of Ipsec features. Bernard remarked that security provided inside the application would scale better than firewalls. János agreed but observed that it is more computationally intensive.
 

Bernard Tuy from RENATER addressed Action 1.17 about NLA-ID allocation and recommended to look at RFC 3100 which specifies /48 for general allocation with exception for some special cases. He then proceeded to his presentation about the IPv6 pilot topology in France. They have regional interconnection from different sides into the G6bone connecting the France Telecom R&D labs, INRIA, 6TAP, Euro-IPv6, WIDE and sfinx, the Paris exchange point. He said that the exchange point is open to commercial ISPs for experimental proposals.  The relation with the exchange point of Euro6ix, an EC project on IPv6 currently in the negotiation phase, was still an open issue. Bernard went on mentioning issues related to addressing of peering points, current pilot status and topology. He said his group was participating in GCAP, Gtpv6, 6NET and Euro6ix. In relation with GÉANT, theGtpv6 addressing scheme interconnection networks still needs to be defined. There are many boxes available from Telebit, Hitachi, Juniper and Cisco, the group needs to discuss what they want to do with them. RENATER plans to have BGP peerings, internal BGP, and IS-IS. Bernard invited other people to express their wish list. Graça remarked that one has to know exactly what people actually want before starting making any plans and said that a further discussion was needed.

Bernard closed his talk by making two announcements. He informed the group about the Plug Test ETSI/G6 bakeoff to be held in Sophia Antipolis on November 19-23, 2001. He then said RENATER was looking for volunteers to participate in video conferencing using v6 multicast. The goal was to cast a regular seminar that is usually attended by a hundred of people. Equipment and software would be provided, including a PIM SM patch available for MLDv2.
 

7. Optical Networking

Victor Reijs from HEAnet and SURFnet introduced the session on optical networking by saying that he was aiming at providing some updated information about the activity and mentioning some ongoing talks with a possible provider of international Lambda connectivity in Europe. Victor could not provide details because the discussion was at a very early stage, but said he was hoping to have more concrete information at the next TF-NGN meeting.
 

Robert Stoy gave a short introduction about Optical Transport Networks (OTN) and G.709. In OTN standardisation the increase of bandwidth is achieved by the adoption of DWDM and Forward Error Correction. Layering in Optical Transport Networks is done in compliance with ITU-T G.872, the protocol that defines optical transport hierarchies. G.709 can be applied at the UNI/NNI of the Optical Transport Network and defines the interface between OTNs to be used at sub-network level of the optical network. There were some questions about Cisco plans of providing G.709 to which Graça replied that Cisco is participating into some projects on UNI with Siemens and Marconi, but the time frame for prototyping was not known exactly. She then agreed to send people a pointer to a list of a special demonstration of G.709 interfaces developed in the project.
 

Victor took again the floor with the aim of brainstorming on Lambda switching and providing an overview of things likely to happen in the near future. He mentioned the Lambda workshop that was held in Amsterdam in September. Victor said all presentations are available at http://www.terena.nl/conf/lambda/ and he specifically recalled an interesting one from Global crossing. He said one of the main things that become clear was Bill St Arnauld idea to consume bandwidth by implementing radically new application (like e. g. a link considered as a storage device), and the related one about user-controlled bandwidth. Victor thinks that real Lambda switching, meaning by this - real color into my machine-, is very far away: the only standards at the moment regard ITU-T and GRID, there are no standards for Lambda interoperability regarding laser modulation and things like spectrum width, power levels, bandwidth (OC-48, OC-192, etc), dispersion management. Victor`s presentation also included an examination of possible set ups: optical switch (switching/changing colors, is protocol transparent, but not really available at the moment), router/switch (transponding colors, not protocol transparent), cross bar (transponding colors, protocol transparent). He said Canarie will start with cross bar, the simplest one.
 
Victor then expressed some of his ideas. He said users are not interested in Lambdas as such, but in having more bandwidth. Instead of using the term Lambda he prefers to use the expression virtual path, no matter if it is a Lambda or something else. There are issues about how Lambdas are going to be managed, whether by the user or desk-to-desk, computer center-to-computer center, or even GPOP-to-GPOP. It is essential to assess which option would be feasible. In comparison with ATM, in terms of capacity increase rate, WDM presents almost the same quantum leap capacity (4x, 16x) in going from 622Mbps to 2.5 and up to 40 Gbps in some year-time. Another consideration was not to expect multiple Lambda interfaces, but multiple GigE/SONET interfaces and no reuse of unused bandwidth. It is expected that applications will most likely look at intra-domain stuff. In fact GMPLS and OBGP are both starting at intra-domain and will need more development to go into the inter domain switching.

Graça remarked that user management bandwidth service is related with the GRID and said it would be nice to liaise with people supporting GRID applications and watch at what they are doing, especially to understand how much bandwidth and granularity they need. Simon pointed out a risk that customers claim needs for dedicated links with infinite bandwidth to access all needed resources and do not want to accept to use existing links which are only marginally loaded. Dimitrios invited the group to look at the issue from a user perspective and take MPLS as it is defined today as an example. The main problems are related to administrative issues between different domains. Mauro was wondering if it is advisable to go back to end-to-end model. He said it is useful but should be done in a different way. It would be useful to switch Lambdas according to load, he does not believe in user managed bandwidth. Jürgen remarked that in their Gigabit test bed in Germany it was difficult to find applications requiring such a high amounts of bandwidth. The most important requirement was that customers need very good quality end-to-end, and this should be the ultimate goal for lambda switching. Tim said in the UK they have a GRID project showing that the problems pointed out by Dimitrios are a critical issues. The real stumbling block is inter-domain administration.
 

8. QoS/SLA monitoring

Victor Reijs mentioned that the GÉANT deliverable 9.4 was available. A draft would be sent to the list. The report shows that one-way delay without GPS is possible, and Victor does not need to work on GPS, he is really looking at different ways of measuring at the user side. He asked who wants to investigate. Roberto encouraged more people in the group to actively take the invitation from Victor to work on the measurement activity and Lada said that Sven Ubik and other people at CESNET have worked on that. He would contact them.

Action 5.4. Lada to invite Sven Ubik and his colleagues in Prague to get involved in the measurement activity.

Victor added that the IPPM measurement protocol has now been made independent of one-way delay; now it can monitor 1-2 ways and make active measurement easier. Infrastructure specifications have been put in the deliverable. The infrastructure should be flexible and able to support different tests, handle different types of granularity and support both active and passive measurement, security, multi platform etc. Tim said a considerable amount of monitoring information has been gathered in the Data Grid project from nine monitoring points in the UK. He will get in touch with them.
 

9. Premium IP Service

Mauro Campanella from Garr informed the group that in Italy they had started making tests on Garr-G. The Garr-G pilot is now operational. It took long time to have it up because, among other tests, they performed a layer2 validation check and monitored SDH/SONET variables with SNMP at both ends of the links.

Mauro provided a summary about the premium IP activity. Traffic between Italy and France was being tested. He said when you receive something from a trusted domain, like GÉANT, you do not need to police at the border. One should limit the total premium capacity when assigning services to users to about 5% of each core link. He gave some examples and said that all details are provided in the deliverable.

There was a discussion about how to police illegal traffic, Mauro explained that there are two available choices: either to discard the packets, or to re-mark them to Best Effort. Mauro said the ID draft ietf-diffserv-new-term-04.txt recommends re-marking and his proposal was in agreement with it, but he wanted this to be discussed by the group. He acknowledged that re-marking is more complex. Jürgen said it could be better not to re-mark in the beginning and introduce it later on when the service is more trusted. Dimitrios noticed that re-marking makes end-to-end debugging easier. For Tiziana debugging is more complicated when you re-mark, and end-to-end monitoring becomes more complicated; dropping is better from a statistical point of view. There was a poll and 14 people were in favour of dropping, 10 of re-marking. Mauro concluded that it is more convenient to drop in the first stage and to re-mark later on. Simon asked whether one should use EF DSCP at all, and in that case what to choose between DSCP 40 and 46. Larry suggested using the most granular one. Mauro said one should aim at using 46. Other remarks were that there is no need of scheduling and classifying on all routers of each domain; a limited set might be sufficient if you have ACL (re-marking) on all interfaces of the selected routers. A simple solution would be to allow only one single DSCP value.  A list of hardware capability needs to be compiled.
 

Afrodite Sevasti from GRnet started with a comment to a discussion raised during the previous speech on SLA/SLS. She said that since one does not have shaping at the ingress one should start monitoring the service performance inside the domain to which the SLA applies. She then discussed SLA definition between two peers. In practice one has a chain of SLA/SLS between domains, the strategy is to configure SLA/SLS associations manually in order to provide end-to-end service. In order to fulfill it end-to-end a particular service instance must made known to all domains involved and therefore there should be some central debugging entity. There should be an SLA between GÉANT and the NRENs. The next step would be to establish an SLA between end users and end-user domains, to define mechanisms for SLA negotiations and then mechanisms to establish an end-to-end SLA chain. The administrative/legal part of the SLA specifies parameters like time duration, response time, help desk, price, etc. The SLS is the technical specification part of the SLA. Its scope is defined in RFC 2475 that also specifies where packets are entering and leaving a domain, flow description, performance guarantees, and other parameters. A template is available from the Sequin project.
 

10. Date of Next Meeting

- The 6th TF-NGN meeting will take place at CERN in Geneva, Switzerland on January 10 & 11 2002.

- The 7th TF-NGN meeting will take place at the University of Southampton in Southampton, United Kingdom on April 18 & 19 2002.
 

11. Any other business

Wim Sjouw from the University of Amsterdam made an extra agenda presentation about some experience with over-provisioned networks where four times 450 UDP traffic flows were injected. Results of statistical tests show that even if the network is loaded 50% there is a significant increase in RTT. Details are provided on his presentation available via http://www.terena.nl/task-forces/tf-ngn/presentations5.html.
 

12. Actions from last meeting

1.13 After the IP premium deliverable has been agreed Robert, Tiziana and Lada to refine the work plan for QoS & Multicast.
Robert made a presentation at the meeting in Athens and estimated to work out a plan within three months.
 - Superseded by new Action 5.1

1.17 Bernard to send to mailing list a format being used to distribute IPv6addresses to customers.
 - DONE

1.18 Tiziana. Mauro, Roberto, Hervé to finalise discussion about loan of equipment. RIPE boxes will be installed on Garr-G.
Substantially done, but Tiziana is interested in continue the discussion, by involving Juniper Networks.
 - Superseded by new Action 5.2

2.1 the MPLS group to check whether MPLS traffic should be tested in  isolation or mixed with BE.
Crihan lacks man power on this activity. All group members should discuss how to tackle this problem.
- Superseded by new Action 5.3

3.4 Tijany to report on the usage of the new generation RIPE measurement boxes.
- Ongoing

4.1 Roberto to post SEQUIN deliverable D2.1 to the email distribution list.
- DONE

4.2 Simon and Victor to work on specification of the monitoring system for premium IP service.
- Ongoing

4.3 GRNET to produce initial thoughts on SLA/SLSs.
- DONE

4.4 Mauro to continue the discussion about the issues of traffic shaping on the TF-NGN email distribution list.
- DONE

4.5 Mauro to take into account comments from people doing experimental work on shaping in his draft report to be distributed in 2-3 weeks to the email distribution list.
- DONE

4.6 GARR and GRnet to start performing the baseline tests.
- Ongoing

4.7 Octavio to provide charts about packet loss.
- Dropped

4.8 Simon discuss his ideas about limitation and restriction in the use of AF.
- DONE

4.9 Wim Barbaix to collect proposals from TF-NGN for further interconnections and experiments with ATRIUM.
1 proposal about QoS testing was received. More proposals are welcome
- Ongoing

4.10 Alain and Mauro to follow up with Graça about hardware/software requirements in setting up the Plage and Garr-G testbeds.
The action was related to the Cisco Engine Cards - Graça received some requests and recommended to set up the equipment with the latest cards. The Cisco equipment is ready to be shipped
- DONE

4.11 Roberto, Alain to define the configuration and request the Smartbit testkit.
- DONE

4.12 Graça to mail details of capabilities of Cisco engines to the TF-NGN email distribution list.
Was sent to people individually
- DONE

4.13 Task Force members to send Mauro requirements for experiments on the Garr-G test bed.
No request was received. The test bed is still available until November 2001
- Ongoing

4.14 Jean-Marc to provide information about support for shaping in Juniper routers.
This action regarded shaping per CoS queue, which at the time of the meeting was not supported yet. [Early in November 2001 Jean-Marc announced that Juniper had just released JUNOS 5.1 that provides CoS enhancement such as Strict Priority, full DSCP support, and the expected Transmission Rate Control per forwarding class (shaping per CoS queue)].
- DONE

4.15 Tiziana to evaluate the suitability of Pioneer facilities for shaping experiments.
Decided to test other equipment and asked Juniper and Cisco to use the higher-end equipment. She is interested in having it as soon as it is available.
- DONE

4.16 Tim to collect items on the IPv6deliverable and to draft a paper in the first two weeks of July.
- DONE

4.17 Simon to coordinate with Victor and collecting information for the GÉANT report due end of July.
- DONE

4.18 Victor to send the list an English version of the Polish document on Bandwidth Measurement.
- DONE

4.19 DANTE to prepare the revised GÉANT road map.
- DONE
 

13. Open actions

3.4 Tijany to report on the usage of the new generation RIPE measurement boxes.
- Ongoing

4.2 Simon and Victor to work on specification of the monitoring system for premium IP service.
Ongoing

4.6 GARR and GRnet to start performing the baseline tests.
- Ongoing

4.9 Wim Barbaix to collect proposals from TF-NGN for further interconnections and experiments with ATRIUM.
1 proposal about QoS testing was received. More proposals are welcome
- Ongoing

4.13 Task Force members to send Mauro requirements for experiments on the Garr-G testbed.
No request was received. The test bed is still available until November 2001
- Ongoing

5.1 Robert Stoy to provide detailed work plan for the QoS & Multicast activity before the 6th TF-NGN meeting.

5.2 Tiziana to coordinate discussion about loan of Equipment.

5.3 the TF-NGN group to discuss how to carry over the MPLS activity.
 
5.4 Lada to invite Sven Ubik and his colleagues in Prague to get involved in the measurement activity.