Study and emulavon of IPv6 internet-exchange-based ... - IEEE Xplore

3 downloads 11514 Views 258KB Size Report
Internet, present in every host and router, and ... new IPv6-based Internet in order to accommo- date the ..... them as the best route (by applying the standard.
IPV6: BASIS FOR THE NEXT-GENERATION NETWORKS

Study and Emulation of IPv6 InternetExchange-Based Addressing Models David Fernández and Tomás de Miguel, Universidad Politécnica de Madrid Fermín Galán, Agora Systems S.A

ABSTRACT This article presents the first results of our study and experimentation with the new IPv6 IX-based address assignment model and its deployment in IXs where peering is organized around route servers. An IX model that identifies where the new IX customers can be located and how services like provider choice or multihoming can be offered is proposed. The article also describes the emulation environment that has been developed to easily create and experiment with complex IX scenarios without involving setup of costly configurations made of several pieces of equipment. This work has been carried out in the context of the European Union funded project Euro6IX.

INTRODUCTION

1

More information about Euro6IX project can be found on project’s Web site: http://www.euro6ix.org

IPv6, the new network protocol for the Internet, was initially designed to solve the important problems related to addressing and routing experienced since the early ’90s in our present IPv4-based network. Furthermore, because IP is a basic and critical piece of the transport infrastructure of the Internet, present in every host and router, and the difficulties and high cost of transitioning the network to a new protocol version, special effort has been made to guarantee the scalability of the new IPv6-based Internet in order to accommodate the billions of new devices and users that will be incorporated in the network in coming years. Scalability was probably the most important requirement in the design of IPv6. This led the designers to choose a strongly hierarchical routing and addressing model in order to tightly control the size of the routing tables in the main routers of the network. Hierarchical assignment of addresses and aggregation are the bases of scalability. However, the routing model proposed for IPv6 makes what should be natural in a deregulated market difficult: changing Internet pro-

IEEE Communications Magazine • January 2004

viders. Due to the hierarchical address assignment used, changing providers implies renumbering the network. This fact is not exclusive to IPv6; it is also true for IPv4 ever since classless interdomain routing (CIDR) was deployed. In order to facilitate change of provider, IPv6 proposes several solutions to make network renumbering easier, as well as a new addressing model centered on Internet exchanges (IXs) that avoids renumbering when changing between providers connected to the same IX. The new IX model is at present roughly defined and not well understood. Further study and experimentation with different alternative configurations are needed. Moreover, its integration with other functional entities currently found in IXs, like route servers, needs to be thoroughly evaluated. This article presents the first results of our study and experimentation with the new IPv6 IX centered addressing model and its deployment in IXs where peering is organized around route servers. We propose an IX model that identifies where the new IX customers can be located and how services like provider choice or multihoming can be offered. We also describe the emulation environment we have developed to easily create and experiment with complex IX scenarios without the need to set up costly configurations integrating several pieces of equipment. This work has been carried out in the context of European Union funded project Euro6IX.1

TRADITIONAL IX ARCHITECTURES Traditional IXs [1] are based on high-speed layer 2 infrastructures, typically build on Ethernet or asynchronous transfer mode (ATM) switches. They are basically used to provide connectivity between the routers each Internet service provider (ISP) brings to the IX (Fig. 1), and frequently do not provide any layer 3 service. The interchange of routing information between ISPs connected to an IX is made directly between border routers by means of the Bor-

0163-6804/04/$20.00 © 2004 IEEE

105

ISP1

ISP2

Routing policies

Routing policies

Routing policies

Routing policies IX ISP4 ISP3 eBGP sessions

■ Figure 1. Classical IX architecture. der Gateway Protocol 4 (BGP4) [2]. When two ISPs reach an agreement to exchange traffic, they establish a BGP session between their border routers; through it they announce the prefixes accessible inside its autonomous system (AS) and other routes they have learned through peerings with other ASs. Routes received from a peering neighbor are always filtered. First of all, it is natural to use interdomain routing to filter routes in order to comply with the routing policies of an AS. Only the “policy-compliant” routes are redistributed, first to internal BGP neighbors and later to other ASs. Moreover, filtering is done for security and stability reasons, to avoid instabilities due to configuration errors in border routers. Connecting to an IX does not necessary imply peering with all the ISPs present in it. As an example, Fig. 1 shows an IX with four ISPs connected and without establishing a peering relationship between ISP1 and ISP4. A complete mesh of peering sessions in an IX could be difficult to manage if the number of participants is high. Each router has to keep a high number of peering sessions and configure different sets of filters for each session. Moreover, if a new participant comes to the IX, almost all routers will have to modify their configurations to peer with it. Therefore, the scalability of full mesh IX configurations is compromised.

PEERING THROUGH A ROUTE SERVER In order to solve the scalability problem, a new system called the route server (RS) was proposed [3]. An RS centralizes the interchange of routes between participants in an IX. Instead of peering between them, providers peer only with the RS, as shown in Fig. 2. Besides, the route filtering function is also moved to the RS, giving rise

106

to an “intelligent” IX, now including level 3 services. Basically, an RS is a modified BGP daemon capable of maintaining several “views” of the routing information being exchanged, one for each participating ISP. Instead of directly applying their routing policies, ISPs store them in a common database named a routing policy registry, using a routing policy description language like RPSL [4]. The policies are processed and stored in the RS, where they are used to filter routing updates flowing through each ISP routing view. In this way, each ISP will receive from the RS the same information it would have received if peering directly with other providers and applying the routing policies itself. Route servers do not forward packets among the routers attached to an IX; they just use BGP’s next-hop attribute to cause the traffic to be exchanged directly between routers, even though the routing information is provided by the RS. Route servers simplify and make safer the interconnection among IX participants. While typically global-scale backbone networks avoid their use in order to have tight control over interdomain routing, smaller regional and local ISPs, who often have fewer technical staff or use less restrictive routing policies, benefit from their use as it greatly simplifies the configuration and management of their routers. The RS was one of the first layer 3 functions to be incorporated into IXs. As we discuss in the next section, IPv6 proposes the addition of new addressing capabilities to IXs.

THE IPV6 ROUTING MODEL IPv6 has its origin in problems that have arisen in the Internet since the beginning of the 1990s due to its exponential growing in size. Those problems — basically address shortage and unmanageable routing table size — have their origins in the way addresses were initially assigned. Organizations requesting addresses were given a range of only three different sizes from which to choose: class A (more than 16 million addresses), class B (around 65,000 addresses) or class C (256 addresses), leading to a decoupling between the number of addresses requested and the real number of addresses assigned. Besides, addresses were not assigned following the network topology, thus preventing the possibility of aggregating address ranges and facilitating the scalability of network routing. As a consequence, any address assignment was translated into a new entry in the routing tables of default free zone (DFZ) routers. CIDR [5], deployed in the network since 1995, partially solved the problems by proposing a hierarchical routing and addressing model, where organizations inherit prefixes from their providers. In this way, all organizations being served by a provider can be aggregated in a low number of short prefixes (ideally only one). Most address assignments since CIDR was deployed have followed this scheme. However, although CIDR has appreciably

IEEE Communications Magazine • January 2004

In order to cope with all these ISP1

difficulties IPv6

ISP2

proposes a strictly Route server

hierarchical routing

Routing policies

Routing policies

and addressing model that essentially follows the principles stated in CIDR:

RS tools Routing policies

hierarchical assignment of addresses and

Routing policies Routing policy registry

routing based on ISP4

aggregation.

ISP3 eBGP sessions

■ Figure 2. Peering through a route server. contributed to shifting the Internet to a hierarchical network where aggregation is frequently used, the number of routing exceptions coming from old networks with addresses assigned before CIDR deployment and multihomed sites (i.e., sites connected to two or more different providers and typically exporting their address prefix through all of them) are still a heavy load that will preclude the Internet from continuing its exponential growth in the next years. Nowadays, DFZ routers have to cope with around 120,000 routing entries in their routing tables. That information has to be transmitted and processed in thousands of expensive border routers, equipped with powerful CPUs and a large amount of memory, making the cost of managing interdomain routing very high. Besides, the high number of ASs (more than 11,000), compromises the scalability of BGP, and is the cause of frequent routing instability. In order to cope with all these difficulties IPv6 proposes a strictly hierarchical routing and addressing model [6] that essentially follows the principles stated in CIDR: hierarchical assignment of addresses and routing based on aggregation. Figure 3 shows the shape of this routing model. Roughly, IPv6 addresses assigned to an organization are divided in three parts: • A global routing prefix that defines precisely where the organization is connected • A subnet identifier, assigned following the internal topology of the organization • An interface identifier that identifies a system inside a link The novelty comes from the fact that the hierarchical assignment of global routing prefix-

IEEE Communications Magazine • January 2004

es is compulsory. This means that the prefix an organization receives depends on the provider to which it is connected. For example, in Fig. 3 the prefix assigned to site E must be a sub-prefix of provider P6, and the prefix assigned to P6 must be a sub-prefix of P4. In this way, all addresses assigned to providers and customers below provider P4 can be aggregated in a few prefixes, considerably reducing the routing table size of “big” routers in the DFZ zone. Provider-independent (PI) addresses (i.e., addresses assigned directly to organizations independent of location), which are nowadays the base for IPv4 multihomed sites, no longer exist in IPv6. As a consequence, if a site changes provider, its global prefix must be changed according to its new location in the global topology. Changing the prefix means renumbering the whole site network, normally a difficult and costly task. Fortunately, IPv6 has been developed keeping renumbering in mind, and several autoconfiguration methods have been defined to make network renumbering easier. Another consequence of the IPv6 routing model is that IPv4 multihoming techniques are not allowed. Multihoming in IPv4 is mostly based on the use of PI address prefixes that are announced through all of an organization’s provider connections. New techniques to allow IPv6 multihoming in a scalable way are being investigated [7], as multihoming is a much demanded feature to achieve reliability or load balancing. Although no clear solution has been defined yet, multihoming is a key feature that clearly needs to be worked out for IPv6 to be successful.

107

The use of a proxy router for

P1

P3

all or some class B customers simplifies the IX manage-

X1

X2

P2 P4

ment as it reduces the number of routers involved.

P5 Site A P6 Site D

However, class B customers’ service

Site C

Site B

IX assigned addressing model Site E

choice is limited, due to the fact that all of them share the same routing policy in terms of ISPs selected or multihome configurations.

XI, X2: Exchanges (NAP, FIX, etc) P1, P2, P3, P4: Long-haul providers P5, P6: Regional providers

Provider assigned addressing model

■ Figure 3. The IPv6 routing model. In order to avoid renumbering when changing provider, as well as to allow a geographically restricted way of multihoming, the IPv6 routing model defines a new way of assigning addresses centered on IXs. It basically consists of allocating prefixes to IXs that are later subassigned to organizations connecting to the IX (Fig. 3). This model decouples the address assignment from the connectivity provision. Contrary to the classic provider-assigned addressing model, where customers get addresses and connectivity from the same provider, in this model, the customer gets addresses from the IX and gets connectivity from one or more of the providers peering at the IX. The advantages of this new IX-based aggregation model are: • Customers can change provider without having to renumber their network. • Customers can multihome with two or more providers on the IX without causing scalability problems, because providers do not need to propagate customer-specific prefixes, only the whole prefix assigned to the IX. However, this model is only roughly defined in IPv6 documents. There are a lot of details to be studied and experimented with before it can be deployed (e.g., the exact way customers connect to the IX, the way routing is organized between customers and providers, the business model behind). All these details are being worked out in various contexts, such as in the Euro6IX project. In the next sections the proposals and preliminary results of our work in this area are summarized.

THE IX MODEL In this section the model used for the study of IX-based address assignment for IPv6 is described in detail (Fig. 4). The core component of the IX is a layer 2 fast switching infrastructure that interconnects all layer 3 devices from:

108

Site F

• Long-haul providers, which peer among themselves in the IX and offer transit services to regional providers. • Regional providers, which also peer among themselves and get transit services from long-haul providers. •IX equipment, including an enhanced RS (ERS) that, apart from the general RS functionalities previously described, will include “mediation functions” for the IX customers [8]. IX will also include equipment to support additional services offered by the IX not treated in this article. Traditional ISP customers connect directly to ISP networks, as shown in the figure. They typically inherit their addresses (usually a /48 prefix [9]) from the ranges assigned to their ISPs (one or more /32 prefixes [9]). As the new IPv6 IX model provides the possibility to have customers with addresses inherited from the ranges assigned to the IX (one or more /32 prefixes as well), three types of IX customers have been identified: • Class A, connected through their own router deployed in IX premises operated by them (A1, A2, and A3 in the figure). • Class B, connected through an IX-operated proxy router (PR) shared with other customers (B1, B2, and B3 in the figure). • Class C, connected to an ISP, as traditional ISP customers, but using addresses from the IX range (C1 and C2 in the figure) Note that the distinction between class A and B customers is more administrative than technical, because a set of class B customers together with their proxy router can be considered a class A customer. The use of a proxy router for all or some class B customers simplifies IX management as it reduces the number of routers involved. However, class B customers’ service choice is limited, because all of them share the same routing policy in terms of ISPs selected or multihome con-

IEEE Communications Magazine • January 2004

LH ISP1

IX

Long-haul providers

LH ISP2

LH ISPm Enhanced route server (ERS)

Regional providers

Route server

Other IX services

IX mediator

Proxy router (PR) ISP 1

ISP n A1

A2

A3

Class A customers Traditional customer

B3

B1

B2 Class B customers

C2

C1

Traditional customer

Class C customers

■ Figure 4. The IX model for IX-based address assignment. figurations. A possible solution to this problem could come from the use of virtual routers (i.e., one physical router running several routing daemons in parallel, each giving service to a different customer). The objective of our IX model is to study and experiment with how these different types of IX customers can benefit from the new services the IX assigned addresses allow: • The choice of the provider from which they want to receive service • Change of provider without renumbering • The possibility of multihoming with two or more providers in order to achieve, for example, fault tolerance, load sharing, or more advanced services Initially, we studied the possibility of separating RS and IX mediator functions into two different devices, in order to keep the RS as it is now and improve fault tolerance (if the IX mediator fails, the RS could still provide interconnection service for the ISPs connected to the IX). In this case, the two devices would interchange routing information through a BGP session. However, this alternative shows a serious problem, because the RS (as would any conventional BGP entity) applies the BGP decision algorithm to routes received from the ISPs. For instance, if two or more ISPs advertise the same prefix route, the route server will choose one of them as the best route (by applying the standard BGP algorithm) and will propagate only this one. As a consequence, the mediator and the customers behind it will not receive all the routing information they need.

IEEE Communications Magazine • January 2004

Although the BGP algorithm could be modified in the RS in order to cope with this problem, to simplify the model and concentrate this study on how to offer the services demanded by IX customers, the decision was made to merge the mediation function and RS functionalities. In the model proposed, all provider routers, class A customer routers, and proxy routers servicing class B customers peer with the ERS using BGP. Therefore, the ERS will be responsible for implementing provider routing policies as well as customer choices. Each ISP connected to the IX will have its own public AS number (ASN) that will be included in the routes generated or propagated by it. For the IX private ASNs can be used, for example, by the RS when peering with ISPs. However, note that appropriate filters will have to be set up to remove these private ASNs in the AS_PATH of the routes announced to other ISPs.

EMULATED SCENARIOS Detailed study of the different scenarios that arise from the model presented above is now underway. The precise scenarios for each type of customer are being defined, as well as their emulation using the tool described in the next section. Figure 5 presents the general peering schema used for the scenarios. As mentioned before, all routers from either providers or IX customers peer using external BGP sessions with the ERS.

109

In the model proposed, all

LH ISP

IX

provider routers, class A customer

RPD GP

routers servicing

GP

eB

LH ISP

Route server

eBGP

using BGP. eB

Therefore, the ERS

GP

eBGP

peer with the ERS

IX mediator

ERS

eBGP

Class A/B customers

routers, and proxy class B customers

CD

eB

eBG

P

will be responsible for implementing

ISP

provider routing ISP

policies as well as customer choices. Class C customers

RPD: Routing policy database CD: Customer database ERS: Enhanced route server

■ Figure 5. The BGP peering scheme.

Depending on the service demanded by each customer (stored in the customer database, CD), different information is sent by the IX mediator through the customer’s router BGP session. For instance, in the simpler case where the customer is single-homed, it will only announce a default route with the next-hop field pointing to the selected ISP router. For customers that demand to be multihomed with two ISPs for fault tolerance, the IX mediator could announce a default route to one of the providers, changing to the other in case of failure. Finally, if the customer demands strict control over its routes, it could receive the full BGP tables from the mediator. Although route control is centralized by the ERS, data traffic goes directly between IX routers using the layer 2 switching infrastructure. ERS uses BGP’s next-hop attribute to direct customer traffic to its provider router. If the provider changes, the CD has to be modified, and the ERS must adjust the next-hop attribute according to the change. When the traffic sent by a customer has its destination on a different provider than the one from which it gets service, the traffic will traverse the layer 2 infrastructure twice: once to reach customers provider router and again to reach the router of the next provider on the way to the destination. This is clearly inefficient. It could be solved using redirecting techniques; however, it is not a major problem, as nowadays layer 2 switching devices are powerful enough to cope with that traffic. Besides, security reasons could preclude direct connections between customer and provider routers (e.g., a provider could include layer 2 filters in its router to only allow traffic from routers with which it peers).

110

The main limitation of the scheme proposed is that it only controls the choice of ISPs for customers’ outgoing traffic. This is an intrinsic limitation, because the path followed by incoming traffic must be the same for all customers, as the whole IX prefix (typically a /32) is announced to ISPs. In order to have different paths for each customer, the IX prefix should be announced in a disaggregated way, breaking the hierarchical routing principle. In the case of class C customers, each ISP will have to announce through their peering session with the ERS the sub-prefixes assigned to its class C customers (typically a /48), as well as distribute that prefix inside its intradomain routing. In this case, the IX prefix is distributed in a disaggregated way. However, disaggregation is limited in scope, as each ISP only has to carry its class C customer prefixes.

THE IX EMULATION ENVIRONMENT In order to study and experiment with the IX model previously described, the decision was made to create an IX emulation environment capable of managing complex IX scenarios with the following requirements. It should: • Be able to emulate complex scenarios made of tens of peering routers interchanging an important amount of prefixes, with or without an RS, using a few physical devices (ideally a single PC-based box). • Allow the connection of peers running on external equipment. • Be as transparent as possible. That is, RS and BGP software implementations should be tested as though they will be running in a real scenario. • Be highly configurable and flexible, capable

IEEE Communications Magazine • January 2004

of emulating a broad spectrum of network scenarios. • Be built using open source tools. Basically we were looking for a flexible environment to test RS-based IX scenarios without involving the investment in time and resources needed to create it using real equipment. After considering different alternatives, we decided to base the emulation environment in the User Mode Linux (UML) [10] open source virtualization tool. UML allows running a completely functional Linux kernel as a conventional user process inside a standard Linux kernel. Several such virtual machines can be run concurrently inside the same physical machine. Each one can be configured with the desired number of network interfaces, as well as the topology that interconnects them. Besides, interconnection with external networks through the physical interface of the hosting machine is also possible. The main advantage of UML is its high flexibility and transparency: processes running in UML behave as if they were running in a real scenario. The main disadvantage is the overhead in performance (in terms of CPU usage, physical memory, and storage capability of the hosting physical box) due to the fact that in order to run a process in the emulated scenario, a complete underlying kernel is needed to support it. Although UML is a powerful general-purpose tool, its use is too complex for a user to build scenarios including many virtual machines and complex virtual network topologies. Furthermore, good knowledge of some Linux operating system details (tap devices, UNIX sockets, virtual bridging, etc.) is needed to start an emulated scenario “by hand.” In order to make the use of UML easier for emulating network scenarios, Virtual Network UML (VNUML) has been developed. VNUML is made of two components: a simple and descriptive XML-based language to write a text file with the specification of the desired scenario; and a parser that interprets the language, and builds and runs the scenario described in the file, hiding all the complex UML booting details from the user. Figure 6 shows a simple example topology and its (very simplified) associated VNUML description. As can be seen, the syntax of the language is so easy that it needs no further explanation. To build this scenario the user only needs to run the parser using the specification file as a parameter: no in-depth knowledge about Linux or UML is needed. Moreover, VNUML allows the definition of the commands to be executed inside each virtual machine. For example, applied to our IX case, it is possible to easily define the configuration of the BGP daemons to be started on each node. VNUML has a comprehensive set of characteristics that, for the sake of brevity, cannot be thoroughly described here.2

CONCLUSIONS This article has presented the results of the study and experimentation carried out so far in the context of IX based address assignment for IPv6. Effort has been mainly invested on defining the

IEEE Communications Magazine • January 2004

RS

IPv6 prefix: 3ffe:fff:1::/64

3ffe:fff:1::1 R1

3ffe:fff:1::3

3ffe:fff:1::2 R2

3ffe:ffff:1::1/64 3ffe:ffff:1::2/64 3ffe:ffff:1::3/64

■ Figure 6. A VNUML example. model and the tools to emulate the scenarios, although interesting results coming from the first scenarios being emulated have already been shown. The IX emulation environment has proven to be very flexible and useful in studying the different alternatives, allowing us to propose BGP configurations to offer the services customers directly connected to the IX will demand. Apart from completing the ongoing tasks, future lines of work are being evaluated (some of them already underway): • The new IX model adds important functionalities to the traditional IXs that have to be managed. It is important to define who will manage them and how, and the business model behind it. •Methods and procedures to automate the configuration of the ERS from a central database of customer information should be studied in order to allow each customer receiving service from the IX to have its own routing policies and peculiarities. Use of standard languages like RPSL and already available routing policy registry databases should be evaluated for that purpose. • Study the use of virtual routers to offer differentiated routing policies to class B customers. UML could be useful for this task. • Study how to relieve the limiting of IX customers’ routing policies to outgoing traffic in case of multihoming. Although as mentioned earlier this is a problem intrinsic to the model, some limited distribution of IX sub-prefixes (maybe to a limited group of ISPs or IXs) could improve the behavior of geographically limited multihoming scenarios.

ACKNOWLEDGEMENTS The work presented in this article has been partially funded by the European Union in the context of the European IPv6 Internet Exchanges Backbone (IST Euro6IX) project. We like to thank all the partners involved in the project for their support and cooperative attitude.

REFERENCES [1] I. Guardini et al., “Specification of the Internal Network Architecture of Each IX Point,” Del. D2.1, Euro6IX, July, 2002.

2

A complete description of VNUML and all its functionalities can be found in http://www.dit.upm.es/vnu ml

111

[2] Y. Rekhter and T. Li, “Border Gateway Protocol 4 (BGP4),” RFC 1771, Mar. 1995. [3] R. Govindan et al., “Route Servers for Inter-Domain Routing,” Comp. Nets. and ISDN Sys., vol. 30, 1998, pp. 1157–74. [4] C. Alaettinoglu et al., “Routing Policy Specification Language (RPSL),” RFC 2622, June 1999. [5] Fuller et al., “Classless Inter-Domain Routing (CIDR): An Address Assignment and Aggregation Strategy, RFC 1519, Sept. 1993. [6] R. Hinden, M. O’Dell, and S. Deering, “An IPv6 Aggregatable Global Unicast Address Format,” RFC 2374, July 1998. [7] J. Abley, B. Black, and V. Gill. “Goals for IPv6 Site-Multihoming Architectures.” RFC 3582, Aug. 2003. [8] C. Ralli et al., “Specification of the Backbone Network Architecture,” Del. D2.2, Euro6IX, Jan. 2003. [9] APNIC, ARIN, RIPE NCC, “IPv6 Address Allocation and Assignment Policy,” ripe-267, Jan. 2003. [10] J. Dike, “User Mode Linux,” Proc. 5th Annual Linux Showcase & Conf., Oakland CA, 2001.

BIOGRAPHIES DAVID FERNANDEZ ([email protected]) is associate professor of computer networks at the Telematic Systems Engineering Department of the Technical University of Madrid. He received an M.Sc. degree in 1988 and a Ph.D. in telematics engineering in 1993, both from the Technical University of

112

Madrid. Since 1989 he has actively participated in several national and international research projects focused on formal description techniques, computer-supported cooperative work, and advanced Internet protocols (BTI, Euro6IX). He works as a technical manager in the deployment of IPv6 at the Telecommunications Engineering School. FERMIN GALAN ([email protected]) received his M.Sc. degree in telecommunications from the Technical University of Madrid (UPM), Spain, in 2002, and is working toward his Ph.D. at the same university. In September 2001 he began working on R&D tasks for the Telematic Systems Engineering Department at UPM. Since April 2003 he has worked for Agora Systems. His research interests include IPv6, OS virtualization, Internet signaling protocols, and P2P applications. T OMAS P. DE M IGUEL ([email protected]) has a Ph.D in telecommunications and has been an associate professor at the Telematics Engineering Department at UPM since 1987. His most recent R&D activities have been related to the design of advanced collaborative services, communication networks architectures, communication protocols, and Internet services. He actively participates in many research projects (NICE, LONG or Euro6IX) and has collaborated in the design of a flexible environment to define CSCW services (ISABEL). He works as a technical manager on the deployment of IPv6 at the Telecommunications Engineering School.

The IX emulation environment has proven to be very flexible and useful in order to study the different alternatives, allowing us to propose BGP configurations to offer the services that customers will demand.

IEEE Communications Magazine • January 2004