[ Curriculum Vitae ] [ Research Area ] [ Projects ] [ (Coffee Machine) Collaborations ]
Associate professor in computer science at the University of Strasbourg since 2008.
member of the LSIIT research laboratory (UMR 7005 CNRS-UdS),
research team Scientific Parallel Computing and Imaging (ICPS),
member of the research group TAG on grid computing.
Previously, (2008-2008), PhD student at the INSA of Lyon - LIRIS laboratory, in the DRIM team.
Grid platforms usually are close worlds: non-grid nodes, like light devices, can not be integrated, while they represent a major current trend with very exiting application perspective. Nowadays, data must be gathered before being treated, because the grid lacks of adaptability (to any need) and adapativity (to any new situation).
Grid solutions must be developed in two ways:
Extensibility: More than a basic mean of communication between the Grid and light devices network, concrete collaboration mechanisms must be designed, in order to exploit the advantages of each platform. On one hand, light devices should be able to ask the Grid for more computing capabilities or to solve their inner network problem. On the other hand, the Grid should be able to ask the light devices network to gather more precise data in order to improve the analyze of one detected macro-phenomena.
Reactivity: The Grid should be able to meet the dynamicity of light devices network. It includes the development of very reactive node reserving and deployment mechanisms, able to monitor light devices needs in order to meet their requirements.
NDS is a Web Service designed to support decision-making processes in complex distribution problems needing optimal solutions.
In grid environment, distribution of resources is a key issue to harness the infrastructure power and ensure optimal performances. To reach the optimality, decisions must be taken according to several aspects:
The managed resources: It can be services, data, DBs, tables, and even computers...
The characteristic of the managed resource: Data have different sizes while services have different sizes of input data, computing times, and sizes of output data.
The objective: It can be response delay, load balancing (of CPU or network), or any other specific objective that one wants to achieve.
The topology and condition of the underlying infrastructures: They are both highly heterogeneous and highly dynamic. Moreover, in grid environment, regular developers do not have access to administration-level information such as IP addressing schemes or routing tables. Such knowledge restrictions make the distribution decision-processes very hard. Moreover, applications must be deployable in different environments and thus highly adaptable.
Taking all these aspects into account makes distribution decision-processes a real challenge for administrators and developers.
To provide knowledge on the infrastructure, efficient monitoring solutions, such as MDS and NWS, have been developed. Unfortunately, these solutions are both very expensive in term of resources consumption and hard use because they provide too low-level information: most of the developers do not want to deal with bandwidth or CPU availability in their applications.
The concept of distance is ideal to facilitate decision-making processes because:
It corresponds to real world representation and thus are easy to understand and manipulate.
Simple decisions can be reduce to simple distances comparisons.
When the distances are used to build a graph representing the network, graph theory and algorithms can be used to optimally solve complex problems and make complex decisions quite easily.
The purpose of NDS is to facilitate the decision-making processes by allowing to build and compute distances adapted to the resource characteristics, the objective and the infrastructure together. Its features are:
Uniform access to raw monitoring information provided by any monitoring system, ensuring to make the most of their use.
High-level complex compound metrics, such as Data Transfer Cost or Computation Task Cost, that are closer to developers and administrators concerns.
Complete extensibility: Users can declare their own metrics and distance functions in a very easy way.
Graph algorithms implementation to solve hard problems like k-centers.
Characterization of the computed distances with satisfaction degrees of the classical metric distance properties: nonnegativity, symmetry, triangle inequality...
GUI to visualize the produced distances and graph.
The project started on 1st of January 2006. It proposes to join efforts of two research groups, a French research group at INSA Lyon and an Austrian research group at the University of Vienna, in order to address new challenges in the development of technology for pervasive information systems. We aim at bringing together the competencies of the French group in data replication management, and the expertise and ongoing research activities of the Austrian group in mobility modeling for mobile distributed computing scenarios. Based on mobility models and on the prediction of future mobile behavior, methods and architectures for proactive data replication management will be developed for mobile users. Based on the technology to be developed, the responsiveness of pervasive information systems will be increased for mobile users, which, in turn, is expected to improve the acceptance of a variety of multimedia based services and applications using content delivery networks.
People:
From Universität Wien, Institut für verteilte und multimediale Systeme: Karin A. Hummel, Wilfried N. Gansterer, Andreas Janecek, Andrea Hess.
From France: Jean-Marc Pierson, Georges Da Costa, Julien Gossa
Publication:
"Proactive
Replica Placement Using Mobility Prediction"
In the last years, several distributed system paradigms have emerged, aiming at the share of a large amount of resources among a large number of users. The Virtual Organizations concept in a multi-institutional context presents obviously numerous opportunities for business applications as well as for scientific ones. Nevertheless, the current trends seem to lead to several independent specialized grids in opposition to the early vision of one generic world wide grid. In such a context, large scale heterogeneous databases must be seen as several specialized databases homed in different grids. The challenge is to allow users of these grids to access all databases in the same efficiency that if they are supported by the same grid. Our proposal is an overlay software architecture built on top of grid middlewares which acts on behalf of the user to facilitate the use of inter-grid databases. The proposed mediation solution deal with database heterogeneity and access rights management.
This collaboration is with Nadia Bennani, Ny Haingo Andrianarisoa, and Sabrina Zefouni from the LIRIS's DRIM Team.
Publications:
"A
User-Profile-Oriented Mediation Architecture for Very Large DataBases
in Dynamic Inter-Grid Context"
"Self
Organizing Inter-Grids Virtual Organization"
The Grid is an emerging solution for sharing resources through a network. It is meant to manage heterogeneous resources in world-scale multi-institutional networks. Grid resources monitoring and network monitoring are very active research areas with actually efficient solutions. Unfortunately, these solutions are limited in term of analysis of the gathered data. Our proposition is to use Data Warehouse (DW) and On-Line Analytical Processing (OLAP) technologies on Grid monitoring information. This allows new complex analysis of crucial importance for Grid users’ everyday tasks. (Un)fortunately, the implementation raises several challenging issues.
This collaboration was with Sandro Bimonte, former member of the LIRIS' Systèmes d'Information Spatio-Temporels & Entreposage Team.
Publication:
"GR-OLAP:
On Line Analytical Processing of GRid monitoring information"
We have proposed an adaptation of personal firewalls technologies to pervasive environment. Personal firewalls have been quickly appreciated by end-users as an effective security solution for every day life. To meet pervasive environment requirements, our proposal is meant to be nonintrusive, to integrate contextual aspects and to be used by end-users without depth knowledge of security purposes. Our contribution on pervasive security concerns is twofold. First we address the problem of variety of computing devices uses and mobility by enriching traditional firewalls rule-sets with contextual aspects. Second we address the problem of security policy configuration through collaborative techniques. We did not aimed at defining a complete security architecture but rather focuses on relevant new principles.
This collaboration was with Romuald Thion, former member of the LIRIS' Systèmes d'Information Pervasif Team.
Publication:
"Gestion
de la sécurité des utilisateurs mobiles : une
adaptation du filtrage de paquet"