The success of the Internet results from its capacity to connect in a transparent way a large number of users. Its evolution will be also conditioned by the ability to interconnect federation of high speed access networks of different technologies. In particular, wireless technologies will be increasingly present in our daily professional and social activities; in parallel, high speed wired accesses spread everywhere. Population of subscribers will be able to communicate using wired or wireless access technologies via complex interoperable architectures. There is currently no tool able to tackle performances of end-to-end services provided by heterogeneous network infrastructures, from the wired or wireless access points (ADSL, Wifi, GPRS, UMTS) to the service platforms. This constitutes an undeniable barrier to the deployment of these new technologies at a lower cost and with a controlled quality of service.
Goals.

AVIPS project handles these problems according to three main objectives:

• the first aims at the design and the development of a software tool for end-to-end simulation of systems constituted of heterogeneous networks and service platforms. The proposed modelling will cover not only the interconnection of heterogeneous networks but also the behaviour of the service platforms.
• the second aims at the modelling of the traffic resulting from subscribers connecting to these services using heterogeneous network technologies. In particular, a major concern is the creation of traffic matrices corresponding to the traffic generated by different categories of subscribers when connecting to shared services with different wired and wireless media via a common core IP-MPLS network.
• the third deals with network planning problems, and resource optimisation problems related to the dynamic adaptation of the system in case of equipment breakdowns, congestion or performance degradations. These policies are on several levels since they relate to:
• the core network (IP and MPLS) and its traffic engineering via LSPs,
• the dynamic assignment policies performed by the DNSs making it possible to ensure on the one hand the continuity of the service in the event of breakdown of a platform and on the other hand a better quality of service according to the load,
• the dynamic resource management inside the service platforms (task mapping, load-balancing, service virtualisation engine).

The prototype developed in the project will allow decision support at the R&D level and design of network architectures and their dimensioning at the operational level. While being based on predictive information contributing to the decision, AVIPS will allow:

• the validation of new architectural concepts (horizontal versus vertical),
• the deployment of new platforms or the extension of the current ones,
• the evaluation of the end-to-end performances of heterogeneous systems,
• an adaptive resource management making it possible to optimise the robustness of the system and to supply an improved end-to-end quality of service for heterogeneous users.

Accordingly, the AVIPS project corresponds completely to the orientation of topic 1 "Mobility and Interoperability" of the call, and partly to the orientation of topic 2 "Very High Speed", since it deals with multiple high speed access networks.
Partner skills.

QoS Design: expert in the design and development of network simulation and planning software, with a consistent software suite called NEST
SFR: strong operational expertise as a Mobile operator, knowledge of operational needs, traffic and equipment characterizations. SFR will be a major partner for the technical validation and industrial valuation.
ENST: expert in the field of the quality of service, the interconnection of heterogeneous architectures and network modelling.
LAAS-CNRS: strong skills in stochastic traffic modelling and optimisation.
IndexMultimedia: operational expertise in the service deployments, the management of interoperable platforms and the analysis of user traffics on a worldwide scale.

Project impact. The Software developed in the AVIPS project will integrate a significant evolution within the domain of planning tools for convergence networks and for the simulation of heterogeneous traffics and service platforms in large scale networks. QoS Design considers a broad marketing of the performance evaluation and optimisation environment which will be developed in the course (and after) of the AVIPS project. The "targeted" customers are network operators, virtual operators, service suppliers, integrators and ISP/ASP. The environment will be capable to bring an effective support for the design of next generation service architectures with guaranteed QoS and optimised costs.

• Design and development of a hybrid modeling multi-purpose tool based on a performance evaluation tool (DHS) for IP Diffserv networks.
• Analytic modeling of routers and packet scheduling.
  
• Multimedia traffic sources modeling for MPLS Diffserv networks:
  • Poisson process based traffic sources
  • Audio (IPP [on-off] model using G711, G726 & G729 codecs)
    
  • Video (MMPP-M model using H261 codec)
    
  • Video (M/G/∞ model using M-JPEG,
  • MPEG1[GOP] codecs)
    
  • Data (Telnet, FTP, SNTMP, NNTP, lognormal, pareto, Poisson...)
    
  • TCP (real TCP protocol - Reno and others)
    
  • HTTP (On-Off combinations, HTTP 1.0 and HTTP 1.1)

The aims of this project was the following:

• Design of a complete analytical model aimed at describing interconnexions between IP Autonomous Systems and an MPLS core network.


• Modeling of the optimization problem associated to the LSP creation process:
  • Regular IP flow routing (best effort) using OSPF
• Analysis of several algorithms for LSPs optimal routing (greedy algorithms, multi-agent methods)
  
• Optimal LSPs rerouting for resilience (for protected LSPs)
  
• Infrastructures dimensionning according to flow constraints (predicted traffic variations, SLA constraints...)
  
• Topology modeling: impact on SDH

Project Partners: Alinka (Prologue Software) and Delta Partners (Anite).

This projects aimed at designing Internet services to be executed on a cluster of processors for heavy industrial applications such as the NetQUAD software. The Clustering architecture is used because of its abilities to be modular, extendable and resource allocatable. A Web portal can be set up for a company to give personalized access for dedicated applications allowing performance evaluation, optimization and simulation of telecommunication networks using the NetQUAD software.

Building such a ASP (Application Service Provider) system requires a software environment for cluster management: system administration, resource management with QoS guarantees, and web access.

The NetQUAD's performance evaluation kernel has been parallelized on the cluster within the final demonstration software.

Project partners: Trinity College of Dublin, CS Telecom, PR-ES Italia, Delta Partners and LAAS-CNRS.

The goal of the project was to design and develop a distributed hybrid simulator for modeling and performance evaluation of large telecommunication networks.

Modeling is based on differential equations for every network flow of every network component (router interfaces, ports). These differential equations are either analytically solved for theoretically known model or numerically solved using event-driven simulations.
This is a completely new approach that gives very accurate results for large scale networks that have thus far been tested (more than 800 nodes). This approach is based on a two-level parallel algorithm. The first level is the set of nodes in the network, and the second one is the set of simulation samples. Our software has been tested through a web interface on the parallel computer IBM SP at the Trinity University of Dublin. Demonstration of the software were given at the Telecom'99 in October 1999 at Geneva.

LAAS-CNRS (France) researchers worked with Anite Telecom on a planning software project for BT Wholesale (see references in article La Depêche du Midi).

Differential models that have been developped at the LAAS for many years (Press) are widely used by British Telecom's planning teams (BT WholeSale Project).

Long research and research valorisation processes made the development of this modeling tool possible. However, besides research valorisation most important are the technical and scientific benefits achieved with the development of such a tool for network simulations. Indeed the dimension of problems solved by these analytical models are enormous.

The whole British Telecom network can be globally processed at once :

• several millions of flows,
• tenth of thousands of links,
• and a thousand of nodes.

This leads to an extreme gain of calculation times and planning algorithms, which is based on our innovative analytic modeling algorithms and new parallelized algorithms for Linux PC clusters and Sun SMP computers. The added value of using these algorithms is evident. Our solution is integrated in the BT Wholesale planning software. The combination of such scientific and technological success's is the first in the world.

A partnership agreement has been has been signed in January 2004 with Sun Microsystem, LAAS-CNRS and QoS Design in the scope of the creation of a new Sun Center of Excellence. The goal of this new Center of Excellence is to develop and promote research in the following areas:

• clusters,
  
• grids,
• policies for dynamic management of grid resources,
• telecommunication application software.

A partnership agreement has been signed with CNRS, LAAS-CNRS and QoS Design. This agreement gives to QoS Design an exclusive licence to use modeling algorithms developed by LAAS-CNRS.