Michael H. Behringer <M.H.Behringer@ DANTE.org.uk>
The first section of this paper outlines the technical issues. It describes the requirements of the national research networks in Europe for a high-speed European infrastructure, as indicated in a survey carried out by DANTE in October 1994. The available technical options for implementing the backbone are discussed and an overview of the proposed implementation strategy is given.
The second section covers non-technical issues in relation to the deployment of a high-speed pan-European backbone. The major problem there is the unavailability of high-speed lines from the PNOs, which, in most of the European countries, still have a monopoly position. The EuroCAIRN project was initiated by the national governments to find a solution for these problems. This section also gives an overview about the EuroCAIRN project, its objectives, strategy and goals.
Keywords: High-Speed Backbone, IP, ATM, Technical Implementation, EuroCAIRN
On the question of line speeds, there was unanimous agreement that at least 34 Mbit/s or multiples are required immediately. The same consensus could be seen on the need for future expansion to higher speeds, beginning with 155 Mbit/s. There was also general agreement on the absolute need for IP as a service[1] on the backbone. Most countries took the view that an ATM service is at least desirable in a future infrastructure, while some considered ATM being essential. While CLNS was considered to be required by approximately half the countries at the time of the survey, in April 1995 none of the regional networks (a regional is defined here as a national research network connecting to the backbone) subscribed to a CLNS service on the successor of EuropaNET.
On the question of which technology should be used for delivering the service, there was no consensus. Five countries are open in respect to this issue, while two stated leased lines as the technology to be used. As there was a general agreement of the importance of an ATM service, it is obvious that ATM is the technology to be used on the backbone.
The need for new applications has been specified as well. In this area there is a general tendency towards multimedia applications such as video conferencing. This kind of application puts special requirements on the provision of a high-speed backbone. Primarily this requires constant bit rate services, which cannot be delivered on today's R&D infrastructure to the extent that is required.
A problem that arose during the work on the survey was the lack of clarity of the border between a technology and a service. For this article, a service is defined as the "towards the national network visible interface between the pan-European backbone and the regional network". The technology used is the implementation on the backbone that allows the delivery of the service; therefore, it is not necessarily visible to the regional. In the case of ATM however, this might coincide, as it then represents the technology on the backbone as well as a possible service delivered to the regional network.
To summarise the result of the survey: the national research networks consider an IP service with line speeds of at least 34 Mbit/s an immediate requirement. An ATM service would at least be desirable. Based on these results, the following section investigates which technical options there are to deliver these services to the national research networks.
Multimedia applications are being deployed on the Internet today. For example Mbone provides users with multicast video and audio sessions over the Internet. Although the quality of especially the video transmission is far from perfect, it is the general attitude of the R&D community that a non-optimal service for many users is better than a perfect service available to only a few users. Thus the "best effort" paradigm of IP seems to meet the demands of the community. Nevertheless with new applications available it will become necessary to deploy services with a constant bit rate. This is not possible with IP.
To summarise, IP over leased lines is able to provide the services needed for the immediate future. However, it is not possible to interconnect ATM networks through an IP network, so that an ATM service cannot be provided on this infrastructure. As IP is a best effort protocol, constant bit rate services can not be delivered.
The ATM Forum[3] is working on a global standard on ATM. The User-Network Interface (UNI) has already been specified [ATM-UNI94], and currently the Forum is working on the Network-Network Interface (NNI). Unless those specifications have been completed and are implemented by all vendors of ATM equipment, the full potential of ATM can not be used internationally, as there are already different vendor's equipment in use today.
Even without the uniform specifications of the NNI, ATM can be used to the extent of Virtual Paths (VP) between two switches today. This amounts to a set of virtual leased lines between a set of end points. Therefore a network could be set up with ATM switching technology, which would add some flexibility to the network with respect to easier changes of the virtual network topology. At the moment, however, it is not possible to make use of all features of ATM such as Switched Virtual Circuits (SVC), as the signalling system is vendor specific. For the provision of an IP service on top of ATM at this stage there are no major benefits in using ATM compared to using leased lines.
The set-up of a European backbone on ATM basis would still have the advantage of being able to provide ATM services at least to the extent of VPs to the regionals, or at a later stage full ATM service. An IP service can be provided on top of the ATM infrastructure, as has been proposed by the Internet community [RFC 1483, RFC 1577]. Other services could be offered on top of ATM as well. The main advantage though is the possibility using the full ATM features such as SVCs and use of multimedia applications easily as soon as they become available. This could then be done by updating the software on the switches already deployed.
The major issue with ATM today is the lack of experience in its deployment. Although there are ATM networks being deployed today, most of them are not considered as production services that have to meet certain service level guarantees. Generally speaking, the development of ATM isn't yet finished, leaving a lot of unknowns for the moment [Laubach94]. This imposes a certain risk on the deployment of a European backbone. In the worst case the ATM Forum might not come to an agreement on a common signalling standard, with the possible consequence of difficult or impossible interoperability in the future. The standardisation process might also take longer than expected, possibly delaying the deployment of an ATM backbone.
None of the national research networks see a need for these techniques. This is because there needs to be an intrinsic benefit for providing IP over e.g. SMDS. In this context "intrinsic" means benefit in operational, financial or technical terms to the operating entity of the backbone. The national research networks would only be able to see the benefit indirectly, as they have no need of and cannot exploit the service underneath the IP layer.
Such an intrinsic benefit from the usage of an underlying protocol could be seen if there were public PNO services on that technique available, so that for example the protocol overhead is hidden from the customer. As outlined above, at the moment there are no higher layer services than leased lines available on a pan-European scale, at least not for the speeds required here. For the operating entity of such a backbone to deploy an underlying protocol has several disadvantages. Firstly, the overhead of the underlying protocol is real bandwidth loss as compared to native IP on leased lines or buying a correspondent service from a PNO; secondly, another protocol adds complexity to the stack, which leads to other complications, like higher fault likelihood and possible interworking problems between the layers.
From the possible candidate technologies, two seem to be suitable for the provision of IP services on a European high-speed R&D backbone. Other services than IP and possibly ATM are not required by the research community in Europe, and will therefore not be considered further in this paper. A pure IP service is straightforward to implement. It is the easiest and most secure way to provide the backbone. IP over ATM at the moment (i.e. with PVCs only) does not show any major advantages compared to native IP, but as this is a new technology it bears certain risks in operational aspects, which make it difficult to provide a service with QoS guarantees. However, as ATM will most likely be needed in the future, there should be a pilot network based on ATM technology.
A B C ---------------+---------+---------+---------+ IP Service |------------------>| | ATM Pilot |-------->| | | ATM Service | |-------->| | IP/ATM Service | | |-------->| ---------------+---------+---------+---------+ Phase 1 Phase 2 Phase 3
Phase 1 starts at the beginning of the project (A) with an IP service and a parallel ATM pilot network. This fulfils the immediate requirements, and furthermore provides the possibility to test the suitability of ATM for the provision of a service network. The start of the ATM pilot network does not need to coincide exactly with the start of the IP service network. The outcome of the ATM trials might be that ATM is not suitable for the purpose needed here, or there might be general problems in the global provision of ATM services in a multivendor environment. In this case, the ATM pilot network can be ceased, and the backbone will consist of an IP service only. In case ATM proves useful to the community, the next step (B) can be undertaken.
In phase 2 the ATM pilot network would be operational and suitable for the provision of a service, with the same or similar service level guarantees as the IP network (see above). The timing of (B) depends on external influences, such as the progress on the ATM development with respect to signalling, as well as internal testing of suitability for the special requirements in this case. During this phase, testing of IP over ATM can be carried out with the possibility of falling back to the IP service network in case of problems. This provides an ideal testing environment for IP over ATM. At a certain point in time it will then become possible to provide the same service level guarantees for IP over ATM as for native IP (C).
In phase 3 there is a backbone based on ATM technology which offers both ATM and IP as a full service to the regional networks. It will be provided on one physical backbone, which offers the best value for money.
In the case that during the ATM pilot phase or at a later stage ATM should, for whatever reason, not prove to be a suitable backbone technology, there is always the option to fall back to the native IP backbone. This ensures that investments into the backbone are not lost completely.
The drawback of this solution is the costs. It is not any certain, whether individual countries can afford a connection to a separate high-speed test network.
* The delay on leased lines has a big impact on the size of buffers in the switch. It is therefore very difficult to extrapolate results from a local environment to a production network.
* The ATM network will be used for IP traffic primarily, which is by its nature bursty. This shape of traffic is difficult to simulate in a test environment. But the impact of this on the buffer sizes in the switching equipment is a very important topic.
It is certainly possible to test the interoperability of different switches in a local environment to a certain extent. But to be able to provide a production service on that equipment requires testing in a real user environment, albeit at a later stage. Therefore lab testing can provide valuable input, but it will not be able to replace field tests completely.
This model can only be deployed when there is reasonable confidence that the ATM part of the network is sufficiently stable. Based on current experience with the configuration of ATM networks that use only PVCs, it seems possible to do that.
The savings in terms of lines are dependent on the topology deployed and on the level of risk one is willing to take. ATM networks only utilising PVCs seem to be reasonably stable, which makes it acceptable to take that risk.
* The pilot backbone can be kept virtually physically separate from the production network, providing maximum security for the production service.
* The bandwidth of 34 Mbit/s will not be used immediately by IP traffic. Therefore it is reasonable to split up some bandwidth for testing purposes. When the testing period is finished, the full bandwidth can be made available for service traffic.
* Costs only have to be paid for one infrastructure. This might lead to possibly high savings compared with the other solutions. The additional costs of TDM equipment should be comparatively small in relation to the overall benefit.
* Given that a TDM splitting of a line can be changed easily, there is considerable flexibility in this approach. So it would be possible to start with 10 Mbit/s for IP and 24 Mbit/s for the ATM pilot; when the IP service bandwidth does not suffice any longer the splitting could be adjusted.
A possible problem here is that some ATM switches cannot operate on random line speeds, but only on certain pre-defined ones, which constrains the flexibility here.
In most European countries PNOs still operate on a monopoly basis. There are only a few exceptions such as the United Kingdom and Finland, where the telecommunications market is liberalised. Thus in most European countries only one provider owns the telecommunications infrastructure and holds a licence to install new lines. This makes the provision of international leased lines through Europe complicated, since in most cases several different PNOs are involved in providing such a line.
The major concern of European PNOs with regard to providing high speed leased lines is the reselling of this bandwidth for carrying voice circuits. Up to bandwidths of 2 Mbps the cost per voice channel is sufficiently unattractive for reselling; 34 Mbps lines though can carry app. 1000 voice circuits, which makes reselling attractive if the cost are considerably less than 17 times the cost of a 2 Mbps circuit. Since international voice telephony is a highly profitable business, the PNOs try to keep possible competitors out of their country either by not selling the bandwidth required or not at reasonable rates. Their monopoly status enables them to do so.
The liberalisation of the European telecommunications market is planned for 1998. By then, the PNO monopolies will have to make the transition to a competitive environment. This change is already taking place in a few countries like the Netherlands, with the effect of lowering the prices for PNO services. The same effect was observed in the United Kingdom when British Telecom lost its status as a monopoly. But most of the countries are only slowly moving in that direction and prices are still very high, compared to the USA. Unless the telecommunications market is fully liberalised it is not certain if high-speed lines will become available under normal conditions.
If the liberalisation does not take effect considerably earlier than 1998, the R&E community in Europe will have to look for other means of obtaining high-speed bandwidth. A possibility would be to agree with PNOs not to resell this bandwidth, and to commit to only use it for research purposes. This would allow the PNOs to offer more favourable conditions since reselling for voice would not be possible. Public funding associated with this might lead to an Acceptable Use Policy (AUP) on the infrastructure, dedicating the lines to research and development traffic for the academic community.
In the USA the AUP model was used for the NSFnet backbone service, which created an artificial boundary on the otherwise seamless Internet and led to a complicated administrative overhead. As the market developed Internet services became generally available, these offered the sort of performance now sought in Europe. An AUP was therefore no longer necessary since there were at least three service providers. The cessation of this AUP service was generally appreciated, even though it is widely acknowledged that the AUP was needed in the beginning. Given the current difficulties in obtaining lines in Europe there might be no other possibility than the provision of another AUP service here, with the well-known problems seen on NSFnet and also on EuropaNET when the AUP was still in place there.
Members of the EuroCAIRN committee were primarily the ministries of research and education of most of the European countries. These bodies fund the international networking facilities in their countries. Given the difficulties in technical as well as political and financial terms such a initiative was deemed necessary.
The key elements of the project were split into three areas:
* Organisation: Work towards a common European policy with regard to R&D networking in the high-speed range. This required a collaboration between all European countries.
* Technical: Promote leading-edge technologies and support the upgrade of current technologies in European networking.
* Industry: Mobilise Europe's IT industry to collaborate with the research community. This extended specifically to the telecommunications market.
The main challenges the project faced were the availability of suitable telecommunications equipment and to get the funding for new infrastructure organised. Both issues need to be addressed on a European level rather than on a country by country basis.
DANTE was subcontracted by the EuroCAIRN committee to produce a recommendation on how to address these issues. DANTE was the logical partner for this activity since it was set up by the national research networks in Europe to organise their international networking. The report produced by DANTE included a broad survey of the requirements of the national networks, technical recommendations as to the suitability of different technologies and a roll out plan for the implementation of a high-speed backbone [EuroCAIRN95]. Financial matters and connectivity issues to the rest of the Internet were part of the survey as well. The general goal of the report was the immediate procurement of a European high-speed backbone.
The Telematics framework provides the basis for resolving the financial issues. However, it is unclear when and where high-speed lines will be made available by the European PNOs. This is still the major issue for establishing a high-speed backbone for the European R&E community.
[EuroCAIRN95] EuroCAIRN, European Research Information Highways, Volume 1 and 2 (Vol 2 = Trans-European High Performance Interconnect for Academic and Industrial Research Networks - A study prepared by DANTE for EuroCAIRN), 1995.
[Laubach94] Mark Laubach: "IP over ATM and the construction of High-Speed Subnet Backbones", Connexions Volume 8, No 7, July 1994
[Møller93] Svend Møller Nielsen, "EuropaNET - contemporary high speed networking", Computer Networks and ISDN Systems (Suppl. 1), September 1993
[RFC 1483] J. Heinanen, "Multiprotocol Encapsulation over ATM Adaptation Layer 5", 07/20/1993.
[RFC 1577] M. Laubach, "Classical IP and ARP over ATM", 01/20/1994.
[2] "User" in this context means a user of the backbone, i.e. a national research network. This paper does specifically not address the end users, but only the perspective of the regionals of the backbone.
[3] For detailed information on the ATM Forum see: http://www.atmforum.com/