Grid`5000
Transcription
Grid`5000
1 Grid'5000 and Grid eXplorer GdX Grid’5000 Grid’5000 & Grid eXplorer Plates-formes de Grilles expérimentales mutualisées à l’échelle nationale ACI GRID & ACI MD Franck Cappello INRIA [email protected] Avec tous les membres participants ACI GRID ACI Masse de Données 2 Grid'5000 and Grid eXplorer Grid’5000 GdX Agenda • Rational • Grid’5000 • Grid eXplorer 1 3 Grid'5000 and Grid eXplorer GdX Grid’5000 Grid experimental platforms rational Grid raises a lot of research issues: Security, Performance, Fault tolerance, Scalability, Load Balancing, Coordination, Message passing, Data storage, Programming, Algorithms, Communication protocols and architecture, Deployment, etc. Theoretical models and simulators cannot capture real life or real size conditions Production platforms have strong difficulties to reproduce experimental conditions How to test and compare? • Fault tolerance protocols • Security mechanisms • Networking protocols • etc. 4 Grid'5000 and Grid eXplorer GdX Grid’5000 Tools for Distributed System Studies To investigate Distributed System issues, we need: 1) Tools (model, simulators, emulators, experi. Platforms) 2) Strong interaction between these research tools Tools for Large Scale Distributed Systems log(cost) Models: Sys, apps, Platforms, conditions math Real systems Real applications “In-lab” platforms Synthetic conditions Key system mecas. Algo, app. kernels Virtual platforms Synthetic conditions simulation emulation Real Real Real Real systems applications platforms conditions log(realism) live systems 2 5 Grid'5000 and Grid eXplorer GdX Grid’5000 Grids are complex dist. Systems. • Systems – nodes, OS, Static – distributed systems mechanisms (resource discovery, storage, scheduling, etc.), – middleware, runtimes, – – Dyn. – – Fault (crash, transient) Workload (multiple users/multiple applications) Heterogeneity (resource diversity, performance) Malicious users/behaviors • Networks – routers, links, topology, Static – protocols, – Theoretical features: synchronous, pseudo synchronous or asynchronous – Disconnection Dyn. – Packet loss – Congestion 6 Grid'5000 and Grid eXplorer Grid’5000 GdX Existing Grid Research Tools • SimGRid and SimGrid2 • Discrete event simulation with trace injection • Originally dedicated to scheduling studies • GridSim • Australian competitor of SimGrid • Dedicated to scheduling (with deadline) • Titech Bricks • Discrete event simulation for scheduling and replication studies • MicroGrid •Emulator with MPI communications •Not dynamic àNo emulator or real life experimental platform àThese tools do not scale (limited to ~100 grid nodes) àThey do not consider the network issues (almost) 3 7 Grid'5000 and Grid eXplorer GdX Grid’5000 We need Grid experimental tools According to the current knowledge: There is no large scale testbed dedicated to Grid experiments à Grid’5000 as a real life system à Grid eXplorer as a large scale emulator log(cost) Grid eXplorer WANinLab Emulab Grid’5000 TERAGrid PlanetLab Naregi Testbed SimGrid MicroGrid Bricks NS, etc. Model Protocol proof log(realism) math simulation emulation live systems Grid'5000 and Grid eXplorer Grid’5000 A New Grid’5000 Generation of Research Tools 8 GdX 4 9 Grid'5000 and Grid eXplorer Grid’5000 NAREGI Middleware Development Infrastructure GdX • Under Procurement – Installation planned in Dec. 2003 – – – – – 3 SMPs, 128 procs total 6 x 128-proc clusters, with different interconnects 1 File Server Multi-gigabit networking to simulate Grid Environment NOT a production system (c.f. TeraGrid) – Mainly geared towards R&D, but could be used partially for experimental production – ~5 Teraflops – To form a Grid with the IMS NAREGI application testbed infrastructure (~ 10 Teraflops, March 2004), and other national centers(voluntary basis) via SuperSINET 10 Grid'5000 and Grid eXplorer GdX Grid’5000 Netbed (Emulab) Comme « Modelnet » Description de la topologie => NS script Utilisation de Dummynet Outils de Mapping routeur_logique => machine physique Utilisation du simulateur NSE (ns emulation) Utilisation de noeuds exterieurs client (40 DSL) 3 Sites 5 11 Grid'5000 and Grid eXplorer GdX Grid’5000 GNET-1 12 Grid'5000 and Grid eXplorer GdX Grid’5000 What do we need for Grid experiments ? 1) Remotely controllable Grid nodes installed in geographically distributed laboratories 2) A « Controllable » and « Monitorable » Network between the Grid nodes à (may be unrealistic in some cases) 3) A middleware infrastructure connecting allowing users to access the Grid sites 4) A playground to prepare experiments 5) A toolkit to deploy, run, monitor, control experiments and collect results We need these components for a nation wide experimental platform 6 13 Grid'5000 and Grid eXplorer GdX Grid’5000 Agenda • Rational • Grid’5000 • Grid eXplorer 14 Grid'5000 and Grid eXplorer Grid’5000 The Grid’5000 Project GdX 1) Building a nation wide experimental platform for Grid researches (like a particle accelerator for the computer scientists) • • • • • 8/10 geographically distributed sites every site hosts a cluster (from 256 CPUs to 1K CPUs) All sites are connected by RENATER (French Res. and Edu. Net.) RENATER hosts probes to trace network load conditions Design and develop a system/middleware environment for safely test and repeat experiments 2) Use the platform for Grid experiments in real life conditions • • • • Address critical issues of Grid system/middleware: • Programming, Scalability, Fault Tolerance, Scheduling Address critical issues of Grid Networking • High performance transport protocols, Qos Port and test applications Investigate original mechanisms • P2P resources discovery, Desktop Grids 7 15 Grid'5000 and Grid eXplorer Grid’5000 Grid’5000 Participants Comité de Pilotage (11) : -Franck Cappello (animateur) -Thierry Priol (Directeur directeur ACI Grid) -Brigitte Plateau (Directrice CS ACI Grid) -Dani Vandrome (Renater) -Frédéric Desprez (Lyon) -Michel Daydé (Toulouse) -Yvon Jégou (Rennes) -Stéphane Lantéri (Sophia) -Raymond Namyst (Bordeaux) -Pascale Primet (Lyon) -Olivier Richard (Grenoble) Comité Technique étendu (28) : Forums: Deployment/exploitation: Franck Cappello (AS1, RTP8) Programming models: Raymond Namyst (AS2, RTP8) Jean-Luc ANTHOINE Jean-Claude Barbet Pierrette Barbaresco Nicolas Capit Eddy Caron Christophe Cérin Olivier Coulaud Georges Da-Costa Yves Denneulin Benjamin Dexheimer Aurélien Dumez Gilles Gallot David Geldreich Sébastien Georget Olivier Gluck Claude Inglebert Julien Leduc Cyrille Martin Jean-Francois Méhaut Jean-Christophe Mignot Thierry Monteil Guillaume Mornet Alain Naud Vincent Néri Gaetan Peaquin Franck Simon Sebastien Varrette Jean-Marc Vincent GdX 16 Grid'5000 and Grid eXplorer GdX Grid’5000 www.grid5000.org 8 17 Grid'5000 and Grid eXplorer GdX Grid’5000 18 Grid'5000 and Grid eXplorer Grid’5000 in November’2004 Grid’5000 GdX Funding: ACI Grid, ACI MD, INRIA, CNRS, Reg. Grid eXplorer Pau Grid 5000 nodes 3 des sites à2004 1 Gbits ~4M€ à forConnexion hardware only (including fund) 9 19 Grid'5000 and Grid eXplorer Planning Grid’5000 Grid’5000 Call for proposals Selection of 7 sites GdX today Call for ACI GRID Expression Funding Of Interest Vendor selection Instal. Final Fisrt First review Demo tests (SC04) Grid’5000 Hardware Switch Proto to Grid’5000 Grid’5000 System/middleware Forum Security Prototypes Control Prototypes Renater connection Grid’5000 Grid’5000 Programming Forum Demo preparation Grid’5000 Experiments Sept03 Nov03 Jan04 March04 Jun/July 04 Spt 04 Oct 04 Nov 04 20 Grid'5000 and Grid eXplorer Grid’5000 Summary of Grid5000 experiments • Networking GdX of Grid’5000 members – End host communication layer (interference with local communications) – High performance long distance protocols (improved TCP) – High Speed Network Emulation • Middleware / OS – – – – – – – Grid’5000 control/access Grid’5000 experiment automation Scheduling / data distribution in Grid Fault tolerance in Grid Resource management Grid SSI OS and Grid I/O Desktop Grid/P2P systems – – – – Component programming for the Grid (Java, Corba) GRID-RPC GRID-MPI Code Coupling • Programming • Applications – Multi-parametric applications (Climate modeling/Functional Genomic) – Large scale experimentation of distributed applications (Electromagnetism, multi-material fluid mechanics, parallel optimization algorithms, CFD, astrophysics – Medical images, Collaborating tools in virtual 3D environment 10 21 Grid'5000 and Grid eXplorer Grid’5000 Grid’5000 Design goal: GdX Experimenting all layers of the Grid software stack Application Programming Environments Algorithms Runtime Middleware Operating System Networking 22 Grid'5000 and Grid eXplorer Grid’5000 Visions utilisateur de Grid’5000 GdX • Comme un grand cluster de clusters – Pas de nécessité de gérer la sécurité (tous les ports sont ouverts entre ressources Grid’5000 à rsh, etc. fonctionnelles, possibilité de tester NFS, etc.) à Tester les exécutifs et les applications sans interférence vis-à-vis de la sécurité (isolation des phénomènes) • Comme un ensemble de clusters dans des domaines d’administration distincts. – Nécessité de gérer la sécurité (au minimum SSH sur tous les nœuds), idéalement :possibilité de sélectionner le middleware de gestion de la sécurité (Globus 2, 3, 4?) à Tester les exécutifs et les applications dans des conditions de sécurité réalistes. 11 23 Grid'5000 and Grid eXplorer Grid’5000 GdX Modes d’utilisation • Partagé (préparation des expériences, taille S) – Pas de ressources dédiées (les utilisateurs se loguent sur les nœuds et utilisent le système par défaut, pas de middleware, etc.) • Réservé (à la Planet-Lab, taille M) – Ressources réservées non coordonnées (les utilisateurs peuvent changer l’OS sur les nœuds qu’ils ont réservés) • Batch (automatique, taille L ou XL) – Les ressources de Grid’000 sont réservées par les utilisateurs et les expériences sont en mode automatique) • Avec planification possible 24 Grid'5000 and Grid eXplorer Grid’5000 Architecture for Grid’5000 prototype (adressage privé) GdX MPLS Routeur RENATER Routeur RENATER vlans RENATER Switch Grid Routeur RENATER Switch/routeur labo controleur Routeur RENATER LABO Firewall/nat Site grid5000 Front-end local Cluster banalisé Adresses privées Controleur = serveur de boot, replicat LDAP?, DNS secondaire, ordonnancement… Front-end local = routes statiques vers grid5000 et default vers Firewall/nat Firewall/nat = 2 interfaces, redirige accès port 22 vers front-end-local 12 25 Grid'5000 and Grid eXplorer Grid’5000 GdX Grid’5000 prototype (Control) user site Users Site 2 (ssh loggin + password) Control Master Control Slave Control commands -rsync (kernel,dist) -orders (boot, reset) Site 1 LAB/Firewall Router Control Slave Test Cluster Firewall/nat Boot server + dhcp Lab’s Network Site 3 Test Cluster System kernels and distributions are downloaded from a boot server. They are uploaded by the users as system images. 26 Grid'5000 and Grid eXplorer Grid’5000 Middleware1(XP)Grid5000 GdX XP: eXPeriments on • Grid’5000 control - Computing Environment deployment (Ka-tools) Experiment automation (security and control) VGrid « mapping a virtual Grid on a real testbed » Monitoring, benchmarking, performance characterization and analysis • Scheduling / distribution - Scheduling : Data transfers, global communications, work stealing,... Data re-distribution in Grid Task distribution and load balancing in heterogeneous Grid Mixed Parallelism (task and data parallelism) Mixing data management and task scheduling Hierarchical and Distributed Scheduling • Fault tolerance - Fault tolerant Grid-RPC (RPC-V) Hierarchical Fault tolerant MPI (MPICH-V) Fault tolerant in data-flow approach (Athapascan) 13 27 Grid'5000 and Grid eXplorer Grid’5000 Middleware2(XP)Grid5000 GdX • Management - AROMA tool : resources management over a Grid of clusters with different classes of services - Mobile agents for open Grid management - Management of Grids and hosted services (security, QoS, monitoring & control, dynamic configuration, …) - Optimization for wide area distributed query processing - Virtualization of data storage on Grids - Automatic Deployment of GridRPC middle tier. - Multiclusters and lightweights Grid resource management (OAR/CIGRI) • Global Computing/P2P Middleware • Grid SSI OS and Grid I/O - - Executing Web Services on Desktop Grid Workers (XtremWeb) Distributing the Coordination in Desktop Grids (XtremWeb) Harnessing Clusters as parallel Workers Probabilistic certification in peer-to-peer systems Large Scale Data Sharing Service based on JXTA (JuxMem) Experimenting management services for textual document in P2P systems Grid file system (NFSG) Grid-aware OS (Kerrighed) Coupling Computational Grid with Reality Center 28 Grid'5000 and Grid eXplorer Grid’5000 Network(XP)Grid5000 GdX • End Host Communication layer - Intelligent Usage of NICs for local and wide area communications - Direct file access over Myrinet : ORFA/NFS and ORFA/LUSTRE • High performance long distance protocols - Alternative Transport for very high speed networks (backpressure) - Differentiated transport with delay control on WAN - Reliable active and non active Multicast - Network Bandwidth optimization in Grid (VTHD++, Paco++). • High Speed Network Emulation - Automatic Deployment of emulated high speed domains - Experiment design for grid flow interactions studies • Grid Networking Layer - Network Resource and QoS on demand - Grid Overlay and Programmable Routers - Measurement Services for network aware middleware 14 29 Grid'5000 and Grid eXplorer Grid’5000 Programming(XP)Grid5000 GdX • Component programming on the grid - ProActive : a JAVA library (parallel, distributed, concurrent computing with security and mobility) - Assessment of scalability, deployment, security and fault tolerance issues - Hierarchical components architecture - PadicoTM/Paco++ combining parallel and distributed computing • RPC Environment - Large scale experimentation of the DIET platform (Distributed Interactive Engineering Toolbox) - Client/Agent/Server model following the GridRPC standard with distributed scheduling agents • MPI Environment - Time sharing Grid resources - Migration over Clusters with heterogeneous high speed networks • Code Coupling - Application coupling with Athapascan - Communication / method invocation rescheduling into ORB (HOMA) - Fluid transfer simulation and geological code with PadicoTM/Paco++ 30 Grid'5000 and Grid eXplorer GdX Grid’5000 Applications(XP)Grid5000 • Multi-parametric applications - ACI GRID-TLSE Project : expertise site for sparse linear algebra - Climate modeling and Global Change - DataGène Project : Functional genomic • Large scale experimentation of distributed message passing applications – Computational Electromagnetism application (JEM3D) – MECAGRID (ACI GRID project, Smash project-team) • Massively parallel computations in multi-material fluid mechanics • Study of numerical algorithms for heterogeneous computing platforms – Grid computing for medical applications (Epidaure project-team) • Interoperable medical image registration grid service – Optimal design of complex systems (Coprin project-team) • Evaluation of parallel optimization algorithms based on interval analysis techniques • Study of load balancing strategies on heterogeneous resources - CFD, astrophysics,… applications - Collaborating tools in virtual 3D environment. - New algorithms (iterative asynchronous methods) 15 31 Grid'5000 and Grid eXplorer Grid’5000 Demos GdX • 5 démos déclanchées à la demande : – – – – – Infrastructure G5k : on/off/reboot, OAR, etc. Expériences réseaux Installations/modifications middleware (Globus) Environnements (DIET, Padico, PM2, Pro-active, XW, etc.) Applications • 3 objectifs – Expérimenter/éprouver l’infrastructure Grid’5000 – Démo pour les extérieurs/présentations/politiques – Démo pour les grandes conférences (SC) • 2 contraintes – Démos très visuelles – Démontrant les capacités de Grid5000 en plus de celles des logiciels 32 Grid'5000 and Grid eXplorer Grid’5000 GdX Other experiments on Grid’5000 • Grid’5000 will be opened to other French Grid researchers (certainly through a selection procedure) in 2005 • Grid’5000 may be connected to the EU CoreGrid testbed and may be used as an experimental platform for CoreGrid researchers (still through a kind of selection procedure) • Meeting before or after September 16-17, gathering ACI MD project members with Grid’5000 participants. 16 33 Grid'5000 and Grid eXplorer GdX Grid’5000 Agenda • Rational • Grid’5000 • Grid eXplorer (GdX) 34 Grid'5000 and Grid eXplorer Grid’5000 Grid’5000 + Grid eXplorer GdX Combining two Grid research instruments log(cost) Grid’5000 Grid eXplorer Relax Real Life Conditions SimGrid MicroGrid Bricks NS, etc. Relax Conditions Reproducibility Model Protocol proof log(realism) math simulation emulation live systems 17 35 Grid'5000 and Grid eXplorer Grid’5000 Grid eXplorer Analogy with physic instruments GdX Experimental Conditions database/ injector 1k Cpu Emulator Measurement units The Cosmotron. This was the first accelerator in the world to send particles to energies in the billion electron volt, or GeV, region, 1953. 36 Grid'5000 and Grid eXplorer Grid’5000 Grid eXplorer 1) Grid experiments under synthetic reproducible conditions Build the instrument: - 2) GdX 1K CPU cluster (may be only 600 depending on the budget) configurable network (Ethernet, Myrinet, others?) configurable OS (kernel, distribution, etc.) A set of emulation/simulation tools (existing + new ones) Multi-users Located/managed by IDRIS Study impact of Scale in Grid/P2P systems - Address critical issues of Grid system/middleware: • Programming, Scalability, Fault Tolerance, Scheduling - Address critical issues of Networking : • High performance transport protocols, Monitoring, Qos - Port and test applications - Investigate original mechanisms • P2P resources discovery, Desktop Grids 18 37 Grid'5000 and Grid eXplorer Grid’5000 A set of sensors in Grid’5000 (Renater) Grid eXplorer Big picture GdX A set of tools for analysis Emulator Core Hardware + Soft: Emulation & Simulation An experimental conditions data base Emulab cluster Validation on Grid’5000 38 Grid'5000 and Grid eXplorer Grid’5000 Grid eXplorer Funding plan Secured: ACI Masse de données: ACI Grid’5000 2004 : INRIA Rocquencourt: INRIA Futurs: 750 155 150 150 Submitted: SESAME Ile de France: ASTRE 2005: 900 K€ TTC 300 K€ TTC K€ K€ K€ K€ GdX TTC TTC TTC TTC à1.2 M€ secured àPossibly 2,4 M€ in total. 19 39 Grid'5000 and Grid eXplorer Grid eXplorer Architecture objectives Grid’5000 GdX Emulation of 1) large scale distributed systems 2) Grid made of clusters of clusters 1 K CPU 64 bits + 64 to 100 Network emulators +Gigabit Ethernet between all nodes (125 Mbits/s) +128 nodes High speed network (Myrinet, Infiniband) +10 to 20 Gbits/s links between cluster partitions +128 nodes High speed network (Infiniband, Myrinet) Infiniband 128 Noeuds Myrinet 128 Noeuds Ethernet Switch 48 ports 10 Gb/s 10 Gb/s Switch 48 ports 1Gb/s 1Gb/s Grid'5000 and Grid eXplorer Grid’5000 ACI Masse de Données DGdX 40 GdX 13 Labs IMAG, ID (UMR 5132), Laboratoire d’Informatique et Distribution, Université de Grenoble LaRIA (UPRES EA 2083), Laboratoire de Recherche en Informatique d’Amiens, Université de Picardie Jules Verne LRI (UMR 8623), Laboratoire de Recherche en Informatique, Université de Paris-sud LAAS-CNRS (UPR 8001), Laboratoire d'Analyse et d'Architecture des Systèmes LORIA (UMR 7503), Laboratoire lorrain de recherche en informatique et ses applications LIP-ENS Lyon (URM 5668), Laboratoire de l'Informatique du Parallélisme LIFL (ESA 8022), Laboratoire d’Informatique Fondamentale de Lille INRIA Sophia Antipolis, UNSA, I3S-CNRS LIP6 (UMR 7606), Laboratoire d'Informatique de Paris 6 LABRI (UMR 5800), Laboratoire Bordelais de Recherche en Informatique IBCP (UMR5086), Institut de Biologie et Chimie des Protéines CEA, Direction des Technologies de l'Information (Saclay) IRISA, Institut de Recherche en Informatique et Systèmes Aléatoires 20 Alain Lecluse (IBCP), 41 Grid'5000 and Grid eXplorer Laurent Baduel, (Projet OASIS, INRIA Alexandre Genoud, (Projet OASIS, INRIA Sophia Grid’5000 Antipolis) Antoine Vernois, (IBCP) Arnaud Contes, (Projet OASIS, INRIA Sophia Antipolis) Aurélien Bouteiller, (LRI), Bénedicte Legrand (LIP6) Brice Goglin (doctorant), (INRIA LIP RESO), Brigitte Rozoy (LRI) Cécile Germain (LRI) Christophe Blanchet, (IBCP) Christophe Cérin, (Amiens, Laria) Christophe Chassot, (LAAS-ENSICA), Colette Johnen (LRI) CongDuc Pham, (LIP) Cyril Randriamaro, (LaRIA) Denis Caromel, (Projet OASIS, INRIA Sophia Antipolis) Eddy Caron, (LIP/ENS Lyon), Emmanuel Jeannot, (Loria) Eric Totel (Supélec Rennes) Fabrice Huet, (Projet OASIS, INRIA Sophia Antipolis) Faycal Bouhaf (DEA)(INRIA LIP RESO), Franck Cappello, (LRI) Françoise Baude, (Projet OASIS, INRIA Sophia Antipolis) Frédéric Desprez, (LIP/INRIA Rhône-Alpes), Frédéric Magniette, (LRI) Gabriel Antoniu, (IRISA/INRIA Rennes), George Bosilca, (LRI) Georges Da Costa, (ID-IMAG), Géraud Krawezik (LRI) Gil Utard, (LaRIA) Gilles Fedak, (LRI) Grégory Mounié (ID-IMAG) Guillaume Auriol, (LAAS-ENSICA), Guillaume Mercier, (LaBRI), Guy Bergère, (LIFL, GrandLarge INRIA Futur) Haiwu He, (LIFL, GrandLarge INRIA Futur) Isaac Scherson, (LIFL, GrandLarge, INRIA Futur) Jens Gustedt (LORIA & INRIA Lorraine) Joffroy Beauquier (LRI) Johanne Cohen, (Loria) Kavé Salamatian (LIP6), Lamine Aouad (LIFL, GrandLarge, INRIA Futur) Sophia Antipolis) Laurent Dairaine, (LAAS) Luc Bougé, (IRISA/ENS Cachan Antenne de Bretagne), Luciana Arantes (LIP6), Ludovic Mé, (Supélec Rennes) Luis Angelo Estefanel, (ID-IMAG) Marin Bertier (LIP6), Mathieu Goutelle, (KIP) Mathieu Jan, (IRISA) Michel Diaz, (LAAS-ENSICA), Michel Koskas (Amiens, Laria) Nicolas Lacorne, (IBCP) Nicolas Larrieu (LAAS-ENSICA), Nicolas Viovy (CEA-DSM-LSCE) Oleg Lodygensky, (LRI) Olivier Richard (ID-IMAG), Olivier Soyez, (LaRIA) Pascal Berthou, (LAAS-ENSICA), Pascale Primet (LIP), Pascale Vicat-Blanc Primet, (INRIA LIP RESO), Patrick Sénac, (LAAS-ENSICA), Philippe d'Anfray (CEA-DTI/SISC), Philippe Gauron, (LRI) Philippe Owezarski (LAAS) Pierre Fraigniaud, (LRI) Pierre Lemarini, (LRI) Pierre Sens (LIP6 / INRIA), Pierre-André Wacrenier, (LaBRI), Raymond Namyst, (LaBRI), Samir Djilali, (LRI) Sébastien Tixeuil (LRI) Serge Petiton, (LIFL, GrandLarge INRIA Futur) Stéphane Vialle (Supélec) Tanguy Pérennou (LAAS) Thierry Gayraud, (LAAS-ENSICA), Thierry Priol, (IRISA) Thomas Hérault, (LRI) Timur Friedman (LIP6) Vincent Danjean, (LaBRI), Vincent Néri (LRI) GdX 42 Grid'5000 and Grid eXplorer GdX Grid’5000 4 Research Topics The 4 research topics and their leaders: -Infrastructure (Hardware + system), -Emulation, -Network, -Applications. Olivier Richard (ID-IMAG) Pierre Sens (LIP6) Pascale Primet (LIP, Inria RESO) Christophe Cérin (Laria) 21 Experiences Grid’5000 I.1 Platform I.2 Virtual Grid I.3 Virt. Techniques I.4 Emul driven Simul I.5 Network emul. I.6 Heterogeneity emul I.7 Communication I.8 Internet Emul. II.1 Engineering tech. II.2 Mobile objects II.3 Fault tolerance II.4 DHT II.5 Data base II.6 Scheduling II.7 Comm. Optimizat. II.8 Data sharing II.9 Uni and multicast II.10 Cellul. automaton II.11 Bioinformatique II.12 P2P storage II.13 Reliability (adversary) II.14 Security II.15 NG. Internet II.16 Grid coupled sys. Grid'5000 and Grid eXplorer Infrastructure X Emulation X X X X X X X Network X X X Application 43 GdX X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X X 44 Grid'5000 and Grid eXplorer Grid’5000 Matériel GdX Budget: 754 000 € TTC (Procédure lancée Lundi 5 Avril). Tranche ferme : environ 400 processeurs (200 nœuds) Tranche conditionnelle 1 : réseau haute performance (Myrinet ou Infiniband) Tranche conditionnelle 2 : augmentation du nombre de processeurs Financements complémentaires : Grid’5000 2004, INRIA ~450 K€ TTC Détails tranche ferme : -Nœuds de calcul : ~400 processeurs (Opteron 2Ghz ou Xeon 64 bits à 3,6 Ghz) -Nœuds réseaux : ~32 nœuds monoprocesseurs (IA32 à 3Ghz) -Nœuds frontaux : 2 nœuds (biprocesseurs Opteron ou Xeon 64 bits, 1 To de disque) -Réseau : Commutateur Gigabit/s (avec bande passante minimale de 125 Mbits par processeurs) 22 45 Grid'5000 and Grid eXplorer GdX Emulation dans GdX Grid’5000 Question de faisabilité technique : exécuter 100 PC virtuels sur 1 PC réel à 10 K PCv sur 100 CPUs (LRI), 100K PCv sur 1K CPUs (GdX), ~1M PCv sur (Grid5000) Non temps réel Etude de Vserver : www.linux-vserver.org (virtual host bsd) Principe pour n PCv: 1 seul noyau (drivers), n root file systems, z processus (z>n) La table des processus « inclut » un champ supplémentaire « PCv » Processus PCv1 1 noyau /vserver/vs1/… Root dir pour chaque Vserver Processus PCv2 /vserver/vs2/… Parties communes Parties spécifiques driver Overhead de Vserver ? Le scheduleur à équité entre les PCv ? 46 Grid'5000 and Grid eXplorer GdX Emulation dans GdX Grid’5000 Question de faisabilité technique : exécuter 100 PC virtuels sur 1 PC réel 10 K PCv sur 100 CPUs (LRI), 100K PCv sur 1K CPUs (GdX), ~1M PCv sur (Grid5000) Non temps réel à Problème : pas assez d’adresses routables à adresses privées NAT ou Vtun : vtun.sourceforge.net (tunnels virtuels par-dessus TCP/IP) Vtun implémente un « Vlan » entre Vservers de PC différents Un Vserver a une adresse IP 192.168.100.2 192.168.100.99 device noyau driver NIC tun0 : 192.168.100.1 Eth0 : 129.175.7.124 192.168.200.2 192.168.200.99 device noyau driver tun0 : 192.168.200.1 Eth0 : 129.175.7.125 NIC Overhead de Vtun ou NAT ? Equité ??? à devra être vérifiée dans GdX 23 47 Grid'5000 and Grid eXplorer GdX Emulation dans GdX Grid’5000 Question de faisabilité technique : exécuter 100 PC virtuels sur 1 PC réel 10 K PCv sur 100 CPUs (LRI), 100K PCv sur 1K CPUs (GdX), ~1M PCv sur (Grid5000) Non temps réel à Problème : Emulation de la topologie (plusieurs tunnels) et des perfs réseaux (Dummynet et NistNet) Un Vserver a une adresse IP 192.168.100.2 DummyNet NistNet Modifiés ?? 192.168.100.99 Où placer DummyNet ou Nistnet ? -> sur un Vserver ??? tun0 : device noyau 192.168.100.1 driver noyau Eth0 : driver 129.175.7.124 NIC NIC Toujours les questions d’overhead et d’équité 48 Grid'5000 and Grid eXplorer Grid’5000 Emulation dans GdX GdX Question de faisabilité technique : exécuter 100 PC virtuels sur 1 PC réel 10 K PCv sur 100 CPUs (LRI), 100K PCv sur 1K CPUs (GdX), ~1M PCv sur (Grid5000) Imaginons que ce soit techniquement faisable : Validation ? (méthode : quels paramètres mesurer, quels tests, etc.) Facteur de ralentissement par rapport au temps réel ? (variation dynamique ? Compréhension de l’impact de chaque brique: Vserver, Vtun, Dummynet/NistNet ?, etc.) 24 49 Grid'5000 and Grid eXplorer Grid’5000 Grid eXplorer and Grid’5000 interactions Design Test/Check GdX Validation under Real Life Conditions Grid’5000 Integration to standard middleware, Deployment, Performance Design Test/Check Validation of Scalability and under Synthetic Conditions Grid eXplorer Scalability, Fault tolerance 50 Grid'5000 and Grid eXplorer GdX Grid’5000 Summary • Researches in Grid and P2P need large scale platforms – • To study protocols, systems, middleware, programming models and applications in real life OR reproducible experimental conditions Grid’5000 and Grid eXplorer - Will be experimental platforms for Grid& Network researchers (like particle accelerator for physicists) A nation wide platform and a large scale emulator Strong relations between this two projects (researchers are the same persons for the two projects) Hardware should be installed by November 2004 Prototypes (security and control) should work for November 2004 Might be opened for experiments in early 2005 25 51 Grid'5000 and Grid eXplorer GdX Grid’5000 Q&A 26