Publications

Dans cette présentation sera d'abord donné un bref aperçu de l'architecture de l'OPN. Ensuite, le concept d'une unité de coordination sera introduit, grâce aux deux exemples, l'E2ECU et l'ENOC, mis en place pour l'OPN, comment elles interagissent, les outils qu'elles utilisent et les procédures qu'elles emploient. Enfin, nous montrerons comment le déploiement à grande échelle de ces circuits n'est pas neutre d'un point de vue opérationnel pour leurs utilisateurs et leurs fournisseurs et comment les solutions adoptées pour l'OPN peuvent être reprises par d'autres projets (ou clients) de ces liens « optiques ».

In this talk, we will give a brief overview of the LHC-OPN architecture. We will then introduce the concept of a Coordination Unit, describe the operations of the two coordination units involved with the OPN (the E2ECU and the ENOC), how they interact, the tools they use and the procedures they follow. Finally we conclude on how the solutions adopted for the OPN can be extended to other projects or customers using a similar private network.

In this talk will be presented the efforts and actions taken by the EGEE project towards network integration in the grid infrastructure. This will mainly include the work of EGEE to provide to the applications and users end-to-end network services (in a broad sense, both network services and operational services) and the expectations of EGEE to enhance the currently provided services.

In this article we propose a description of the networking activities within a grid project: EGEE. These activities are focused on the access to the provided services (resource reservation) and on the operational constraints of such an architecture (operational interface with networks, sla installation, monitoring). Even if the egee project may have some peculiarities, numerous issues are not egee specific, particularly the ones about network. They prefigure a possible evolution of future network usage, not only by egee or other distributed computing projects, but by any potential user.

Dans cet article, nous vous proposons de découvrir les activités réseau d'un projet de grille : EGEE. Ces activités sont centrées autour de l'accès aux services fournis (réservation de ressources) et de la contrainte opérationnelle d'une telle architecture (interface opérationnelle avec les réseaux, institution de SLA, supervision). Même si ce projet présente certaines spécificités, de nombreuses problématiques sont généralisables, notamment celles concernant le réseau. Elles préfigurent de l'évolution possible de l'utilisation des réseaux dans le futur, non pas seulement par EGEE ni même par d'autres projets de calcul distribué, mais par tous les utilisateurs. Après une courte présentation du projet, nous allons détailler quelles sont les activités relatives au réseau dans EGEE. Ensuite, nous décrirons les implications que ce type de projet peuvent avoir sur les réseaux, notamment les réseaux (locaux, métropolitains, régionaux, etc.) entre le réseau national académique (NREN) et le centre de ressources (calcul, stockage). Enfin, nous conclurons.

After a short presentation of the EGEE project, we will present the EGEE networking activites (SA2 - Network Resource Provision and JRA4 - Network Services Development): Institution of SLA between EGEE and NRENs (Applications Requirements gathering, Definition of Services Classes, Operationnal procedures and policies), Operationnal Interface for Trouble Ticket Management, Network performance monitoring, and Bandwidth Advanced Reservation.

Grids generally rely on a complex interconnection of Internet Protocol (IP) domains that offer heterogeneous services and unpredictable performance characteristics, particularly at the local area network/wide area network boundary. The total lack of end-to-end resource control in IP networks is responsible for performance problems that may affect the whole Grid environment. An end-to-end service differentiation architecture that controls heterogeneous communication performance is thus needed. We propose the Equivalent Differentiated Services (EDS) architecture, based on a layer-4 service differentiation solution exploiting a new layer-3 relative DiffServ model. In this paper, we present the EDS packet forwarding principles, the router mechanisms and two adaptive packet marking algorithms. As a proof of concept, we have implemented the EDS architecture in Linux and performed experiments on a transoceanic testbed.

Studies and tools development for applications sensitive to data rates is a very active research field. Tools for measuring the end-to-end performance of a link between two hosts are very important for distributed application performance optimization. Bandwidth evaluation methods aim to provide realistic view of the raw capacity but also of the dynamic behaviour of the interconnection that may be very useful to evaluate the time of bulk data transfer. Existing methods differs according to the measurements strategies and the evaluated metric. This paper analyses the available bandwidth measurements and the total capacity measurements approches. None of the proposed tools, based on these methods permit to evaluate both metrics, while giving an overview of the link topology. An approach using a hop-by-hop packet pair method and a fine analysis of the measurements can provide such informations. The principles of this hop-by-hop measurement of the capacity and available bandwidth method are exposed in this paper. This proposition has been validated in simulation, then implemented in Linux and validated experimentally. We have compared our method with others to define its limits and the potential utilisations on the developed tool.

In this technical report, we describe the structure and organization of the networking code of Linux kernel 2.4.20. This release is the first of the 2.4 branch to support network interrupt mitigation via a mechanism known as NAPI. We describe the main data structures, the sub-IP layer, the IP layer, and two transport layers: TCP and UDP. Neither IPv6 nor SCTP are considered here.

Standard TCP (TCP Reno) is a reliable transport protocol that is designed to perform well in traditional networks. However, several experiments and analyses have shown that this protocol is not suitable for each and every kind of application and environment - e.g. bulk data transfer in high bandwidth, large round trip time networks. In this document, we review and compare different emerging alternatives that try to solve this and other problems.

This article studies the problem of service differentiation in IP networks. The limits of the standard models are analysed and alternative solutions like the proportional differentiation and the non-elevated services are presented. These services are simpler to set up but need adaptation layers to insure correct end-to-end performances. To illustrate these approaches the EDS model from the INRIA/RESO team and the QBSS-LBE model of the Internet2 project are explored and their properties studied practically. This study highlights the benefit of a global approach of Quality of Service in Internet and open many perspectives of research on the transport layer.

Studies and tools development for applications sensitive to data rates is a very active research field for distributed application performance optimization. The research community works to propose tools for measuring the end-to-end performance of a link between two hosts. Delay measurements provide a first approximation but aren't sufficient enough because the delay isn't a relevant metric. A bandwidth evaluation method would give a more realistic view of the raw capacity but also of the dynamic behaviour of the interconnection, when we want to evaluate the transfer time of an amount of data. Among all the existing methods, there are some differences according to the measurements strategies and the evaluated metric. We first describe the available bandwidth measurements and then the total capacity measurements approaches. Among all the presented methods, none of them can evaluate both metrics, while giving an overview of the link topology. By using a hop-by-hop packet pair method, we show that we can provide such informations with a fine analysis of the measurements. In this report, we detail our proposition of a solution for an hop-by-hop measurement of the capacity and available bandwidth. This method has been validated in simulation, then implemented in Linux and validated experimentally. We compare this method with others to define its limits and the future utilisations on the newly developed tool.

The Equivalent Differentiated Services (EDS) provide a trade-off between delay and loss rate, in order to keep the global performances of the DiffServ classes equivalent. Even if the network caracteristics can be well defined, the end-to-end performances are more difficult to extract. In this report, we present a theoretical and practical evaluation of the TCP protocol on an EDS test platform.

The proportional Diffserv (PDS) model is a model for providing service differentiation in IP. It ensures that a class of service obtains a performanc- e (in terms of delay or loss rate) proportional to the performance obtained by an other class, according to a specific coefficient. The WTP and PLR schedulers provide proportional differentiation. They have not been widely tested today because commercials routers do not implement them. In this report we present an implementation in the Linux operating system of these schedulers and show experimental results we obtained with them.