Section: Overall Objectives

Research area

The use of networks for on-demand computing is now gaining in the large Internet, while the optical transport layer extends to the edge (fiber to the home). Enabling ultra high performance machine to machine communications lead then to new bandwidth and network resource sharing paradigms. RESO is investigating several issues such as quality of service, transport protocols, energy efficiency, traffic metrology , traffic modeling and network resource scheduling to deliver the emerging traffic in a timely, efficient, and reliable manner over long distance networks. In particular, RESO focuses of key issues such as :

• where and how to integrate the autonomy required to manage and control high speed networks at large scale?

• how dealing with the cost of resource and network virtualization on communication performance?

• which type of congestion control should be used in the context of a large scale deployment of high speed access networks (fiber to the home)?

• is the grid and high speed networks traffic self similar as the Internet traffic is? What are the critical scales? In which sense is self similarity harmful?

• how to efficiently share the bandwidth in a network that interleaves multimedia applications and computing applications?

• how to improve the interactions of message based communications or interactive traffic and transport layer in wide area networks?

To address some of these issues, our work follows four major research axes :

• Axis 1 : Optimized protocol implementations and networking equipements

• Axis 2 : Quality of Service and Transport layer for Future Networks

• Axis 3: High Speed Network's traffic metrology, analysis and modelling

• Axis 4: Network Services for high demanding applications

A large part of the axis 2 and axis 3 research topics is integrated in the ADR Semantic Networking of the common INRIA Bell Labs laboratory we are animating. The motivation of our research work in the common lab is to build and to exploit the knowledge that comes along with traffic. The goal is to act in a better way and to make better decisions at router and network level. The knowledge that comes part of traffic is what we refer to the “semantics" of traffic. The main topics we are exploring in this research axis of the common laboratory are

• Traffic identification and classification

• Traffic sampling

• Flow analysis

• Flow scheduling

• Sampling-based scheduling

• Flow-based routing

Axis 1 : Optimized protocol implementations and networking equipements

In this research axis we focus on the implementation and on the optimization of the mecanisms and process within networking devices. Since several years, virtualization of the operating system is used in end system to improve security, isolation, reliability and flexibility of the environments. These mechanisms become a must in large scale distributed system. In our research axis1 we explore how these mechanisms can be also adapted and used in data transport networks and specifically in switching and routing equipments.

However, virtualization introduces an overhead which must be integrated to system performance models in order to forecast their behavior. Lot of performance problems on end systems but also on router's data plane have to be studied and solved to make the virtualization approache viable. Investigating these issues is one of the goals in this research axis.

On an other hand, the key enabling factor of new network services is programmability at every level; that is the ability for new software capabilities to self-configure themselves over the network. We explore the concept, "dynamic programming enablers" for dynamic service driven configuration of communication resources.

In this research axis we also explore the integration of context-awareness functionality to address two important issues : reliability of communications and energy comsumption.

This direction is supported mainly by the EU FP7 “Autonomic Internet" project (2008-2010), with the INRIA “Action de Recherche Concertée" untitled "Green-NET" (2008-2010). The ANR HIPCAL (2007-2009) grant helps our studies around network virtualisation.

Axis 2 : Quality of Service and Transport layer for Future Networks

The goal of this axis is to guarantee quality of service in machine/user to machine/user communication while using efficiently the resources of the future networks. The two problems that are tackled here are: i) dynamic bandwidth sharing and congestion control in Future Internet and ii) control and flow management in semantic networks.

In this research axis, we focus on the three following questions:

• 1) which type of congestion control and transport protocol should be used in the context of large scale deployed high speed networks (fiber to the home, for example)?

• 2) how to efficiently share, but also dynamically provision the bandwidth of a network dedicated to computing tasks?

• 3) is the "flow-aware" approach a valuable solution to solve the end-to-end quality of service issue in the very high speed Future Internet?

Axis 3 : High Speed Network's traffic metrology and statistical analysis

Metrology of wide-area computer networks (i.e. the deployment of a series of tools allowing for collecting relevant information regarding the system status), is a discipline recently introduced in the context of networks, that undergoes constant developments. In a nutshell, this activity consists in measuring along time, the nature and the amount of exchanged information between the constituents of a system. It is then a matter of using the collected data to forecast the network load evolution, so as to anticipate congestion, and more widely, to guarantee a certain Quality of Service, optimizing resources usage and protocols design.

From a statistical signal processing viewpoint, collected traces correspond to (multivariate) time series principally characterized by non-properties: non-gaussianity, non-stationarity, non-linearities, absence of a characteristic time scale (scale invariance). Our research activity is undertaking the development of reliable signal analysis tools aimed at identifying these (non-)properties in the specific context of computer network traffic. In the course, we intend to clarify the importance of granularity of measurements.

Another challenge in network metrology is the effectiveness of packet sub-sampling. It means, to collect only a fraction of the overall traffic (supposedly redundant), and to study the possibility of inferring from that partial measurement, the most complete information about the system. Non trivial questions as, which fraction, which sub-sampling rule, adaptativity of this latter, smart sampling, statistical inference, open up a broad scope of investigation.

In this research axis, we focus on the two following questions:

• how does the traffic statistical properties really impact Quality of Service (QoS)?

• how to identify and to classify, in real time, transiting flows, according to a sensible typology?

Within the framework of the common laboratory between INRIA and Alcatel-Lucent, axis, "Semantic networking" brings in a new field of metrology research in RESO.

Axis 4: Network Services for high demanding applications

In strong interaction with the three fundamentals axes, this axis focuses on the application of the solutions to the grid context and on their implementation in a real environment such as the national research instrument Grid5000. Indeed, we believe that the precise structure of future applications and services is difficult to design without building large scale instruments and systems for real use based on real and high-performing hardware. Therefore, in this research axis we develop prototypes and deploy them within the Grid5000 testbed. For example, we design special measurement and routing systems at the edge of each Grid5000 site to explore new approaches or difficult problems alike. Topics that are investigated in this axis are strongly focusing the usage and the evolution of the Grid5000 instrument:

• Studies on the interactions of MPI and transport layer in wide area networks,

• Design, development and evaluation of a dynamic bandwidth provisionning service,

• Studies on the virtual private execution infrastructure concept for grid and cloud computing environments.

• Large scale deployment and evaluation of a high speed network measurement infrastructure.

RESO pursues researches for improving communications in grid environments. Thanks to systematic experiments of the behavior MPI in large scale environment, we merge optimizations of current implementations and propose new optimizations in the communication layers in order to execute more efficiently MPI applications on the Grid. We also study the impact of using TCP protocol for WAN communications (inter-site communications in the grid) and its interactions with MPI applications.

This research direction is mainly supported by FP6 EC-GIN grant, Grid'5000-ALADDIN initiative, CARRIOCAS project , HIPCAL project, JSPS-NEGST project. The DSLAB and OGF-Europe project provide resources.