Overview of grid / cloud research in France M. Daydé Institut des Grilles, CNRS / INS2I June 1 st 2010
Introduction ● « Large scale distributed infrastructures », Grids since 10 years and Clouds nowadays attract considerable attention in France as in other countries ● Attempt for an overview of grid research in France in terms of topics, structuration and trends ● Based both on the work effected within the Grid'5000 community and on the web sites / activity reports of the labs : likely not exhaustive
Research topics
Research topics ● Research adressing large scale distributed computed infrastructures is quite old and have started in the nineties for Grid Computing. Since 2008 Cloud Computing has emerged as a hot topic. Grid (blue) vs Cloud (red) on Google
Researches on Large Scale distributed platforms (large clusters, grids, clouds, sensor networks, …) ● Environment: middleware, efficient use, distributed systems / objects / software infrastructures, large scale data management, security, virtualization, reconfiguration, autonomic management,heterogeneity, ... ● Services : service oriented computing, management, dynamic adaptation, workflows ● Distributed applications: new concepts, languages and tools, context-sensitive applications, mathematical modelling, abstraction ● Objects or software components architectures for distributed intelligent systems
Researches on Large Scale distributed platforms (con't) ● Algorithms and scheduling: distributed algorithms, P2P, large scale computations, combinatorizal and NP-hard problems, ... ● Design of distributed applications, parallel programming, refinement, development and proof of parallel programs, application of formal languages, models ● Approximatively 530 researchers (450 academics, 50 INRIA researchers and 30 CNRS researchers).
Grid / Cloud research topics ● Environment: middleware, languages, libraries, high performance execution support, networks, communications, energy, … ● Management of ressources: virtualization, autonomic administration, workflows..., ● Management of data in large scale distributed systems (clusters, grid and P2P), distributed storage of data ● Scaling, algorithms and scheduling ● High performance networks ● Pervasive grids
Grid / cloud research topics (con't) ● Modelization and utilization of large scale data and computing infrastructures ● Gridification of applications : simulations and numerical algorithms in various areas (environment, electromagnetism, chemical engineering, biology, health, linear algebra, combinatorial problems, …) ● 12% of researchers in computer science are studying large scale distributed infrastructures. Within these 12% around 310 researchers (7%) are working on grids and clouds
Overall structure of Grid Research in France
The Grid'5000 research community ● Most of the grid research in France (3/5 of researchers) is structured around the GRID'5000 national platform ● The ALADDIN-GRID'5000 initiative driven by INRIA ensures the development and the support of the infrastructure. ● The HEMERA initiative also supported by INRIA is in charge of the scientific animation around Grid'5000. It is organized around scientific challenges and working groups.
Grid'5000 nation-wide experimental platform ● 9 french sites: Bordeaux, Grenoble, Lille, Lyon, Nancy, Orsay, Sophia-Antipolis, Toulouse et Rennes connected using RENATER – and one brasilian site (Porto Alegre). ● It provides over 5,000 cores. ● Goal of Grid'5000: nation- wide platform for distributed computing at large scale. ● See http://grid5000.fr
HEREMA ● 21 teams (15 INRIA, the 6 left are in Strasbourg and Toulouse). ● All teams are in labs associated to CNRS (most are computer science labs in INS2I, 2 in INSIS) ● 174 researchers (119 academics, 38 INRIA and 17 CNRS) which is 3/5 of the researchers in grid in France.
Scientific challenges in Hemera Network: 1. Tra ffi c Awareness System: 2. Energy profiling of large scale applications 3. Robustness of large systems in presence of high churn Programming paradigm: 4. Large scale computing for combinatorial optimization problems 5. Scalable Distributed Processing Using the MapReduce Paradigm Application domain specific: 6. Multi-parametric intensive stochastic simulations for hydrogeology 7. Thinking GRID for Electromagnetic Simulation of oversized structures
Working Groups in HEMERA ● Transparent, safe and e ffi cient large scale computing ● Energy E ffi cient Large Scale Experimental Distributed Systems ● Bring Grids Power to Internet-Users thanks to Virtualization Technologies ● E ffi cient exploitation of highly heterogeneous and hierarchical large-scale systems ● E ffi cient management of very large volumes of information for data-intensive applications ● Completing challenging experiments on Grid’5000 ● Modeling Large Scale Systems and Validating their Simulators ● Network metrology and tra ffi c characterization
Research teams involved in HEMERA ● Bordeaux : CEPAGE and RUNTIME (LaBRI). ● Grenoble :MESCAL (LIG). ● Lille : DOLPHIN (LIFL). ● Lyon : GRAAL and RESO (LIP). ● Nantes : ASCOLA (LINA). ● Paris : GRAND-LARGE (LRI) and REGAL (LIP6). ● Rennes : ASAP, KERDATA, MYRIADS, SAGE (IRISA) ● Nice-Sophia : OASIS (I3S). ● Strasbourg : ALGORILLE (LORIA), ICPS (LSIIT). ● Toulouse : ACADIE , APO , ASTRE (IRIT), MINC and MRS (LAAS)
Other Grid Teams ● Team of David Hill (LIMOS, Univ. Clermont). ● AND and CARTOON (LIFC, Univ. Franche-Comté) ● MAP (LIFL, Lille). ● ATLAS-GDD (LINA, Nantes). ● PRV (LIFO, Orléans). ● Groupe Grilles (LAL, Orsay). ● PEQUAN (LIP6, Paris). ● T2I (LMA, Univ. Pau), ● Groupe SysCom (CRESTIC, Univ. Reims). ● MODALIS (I3S, Nice-Sophia). ● Around 124 researchers (119 academic + 4 CNRS + 1 INRIA). ● All labs associated to CNRS except 2 (Pau, Reims)
DOLPHIN (LIFL) MAP (LIFL) GRAND-LARGE (LRI), SysCom (CRESTIC) REGAL (LIP6) Groupe Grilles (LAL) PRV (LIFO) ALGORILLE (LORIA) PEQUAN (LIP6) ICPS (LSIIT) ASAP, KERDATA, AND, CARTOON MYRIADS, SAGE (IRISA) (LIFC) ATLAS-GDD (LINA) ASCOLA (LINA) D. Hill (LIMOS) GRAAL, RESO (LIP) MESCAL (LIG) CEPAGE, RUNTIME (LaBRI) OASIS (I3S) T2I (LMA) MODALIS (I3S) ACADIE , APO , ASTRE (IRIT), MINC, MRS (LAAS)
Some trends in grid / cloud computing Working groups in HEMERA give some of major trends for the next years: ● Models for large scale distributed infrastructures, experimental methodology, reproducibility of experiments ● Security on large scale platforms (cluster, grids, clouds, …) ● Taking into account dynamicity at all levels: virtualization, autonomic management, monitoring, process automation, … ● Mastering scalability ● Energy-aware management of ressources : algorithms, scheduling, monitoring, …
Emerging topics specific to cloud ● Cloud architectures and provisioning ● Grid / Cloud interoperability, private / public cloud interoperability ● Cloud and HPC ● Programming models, applications and middleware suitable for dynamic cloud environments. ● Novel human interfaces and browsers for accessing clouds: interaction of mobile computing, mCommerce and Clouds. ...
Large amounts of data ● Crucial issue: management and exploitation of large amount of data (possibly heterogeneous) – arising from numerical simulations at Terascale (and soon Exascale) – or from various applications in biology, high energy, semantic web, … ● These data are often processed on large scale distributed infrastructures (e.g. data arising from the LHC). ● Storing / indexing them and performing some post-processing for information retrieval or visualization purposes still require efforts and interactions between the various communities (grid / cloud, visualization, large scale databes / data warehouses, ..).
Conclusion ● Well structured grid / clouds research in France: – major initiative (Grid'5000) driven by INRIA – strong involvement of CNRS – Software pervasing from to research to production (DIET, XTREMWEB, OAR, Kadeploy, Proactive, ...) ● Good visibility at international level: – EU : CoreGrid network, forthcoming EU project (PIRE ?) – Cooperations: NAREGI( Japan), DAS-3 (Netherlands), FutureGrid (USA), ... ● Grid'5000 is a pionneering platform (flexibility, reproducibility of experiments, system image deployment) ● Interactions production / research grids ● Interaction grid / cloud / HPC / large scale data management
Recommend
More recommend
