Grid introduction

Enabling Grids for E-sciencE Introduction to GRID computing Introduction GRID Tutorial Jules Wolfrat SARA www.eu-egee.org INFSO-RI-508833 Definition of Grid Enabling Grids for E-sciencE • From an EU brochure: – It doesn’t matter if your team is modeling the Earth’s atmosphere, designing cars, creating animated films or finding new medicines, the basic principle is the same: your Grid supplies all the computing power, software, data and knowledge you need in one integrated package, and helps project teams work more closely together • The analogy with the power grid: – Like you can plug in anywhere to the power grid without knowing where your energy is coming from you can plug into the grid without knowing where your (computing) resources are coming from. INFSO-RI-508833 Grid Tutorial, Groningen, September 2006 2 History (1) Enabling Grids for E-sciencE • From a news item in 1991 – “Smarr describes the metacomputer as a network of heterogeneous, computational resources linked by software in such a way that they can be used as easily as a personal computer” – So the concept was introduced already in the early 90s, known as metacomputing. – Motivation was the emergence of computer networks. INFSO-RI-508833 Grid Tutorial, Groningen, September 2006 3 Example (1) Enabling Grids for E-sciencE Following is an example of the kind of initiatives started in those years from close by: In 1996 a project was started in Amsterdam: The Amsterdam Metacomputing project is an ongoing effort from the University of Amsterdam (UvA), the Free University (VU) and "Academic Computing Services Amsterdam" (SARA) to develop a Metacomputer environment on the Amsterdam campus. Important components of this environment will be: automatic distribution and monitoring of jobs over a network of computer systems, uniform access to files of other users from each place to work and to each computer system incorporated in the environment, distributed storage of data on various fileservers, automatic backup, migration and archiving, general availability of both commercial and public domain software on software servers, and a minimum of system management tasks. In this way scientists will be able to devote all of their time to their actual task: science. INFSO-RI-508833 Grid Tutorial, Groningen, September 2006 4 Example (2) Enabling Grids for E-sciencE An extensive package of services will gradually be implemented and finally include the following components: • fileservers and distributed, transparent file-systems; • backup, migration and archiving services; • batch-queueing systems, designed for efficient use of local systems, and if desired, of computational servers supplied by SARA; • public domain and specialist (commercial) software servers. All components will be accessible from the scientist's desktop. A client-server architecture will play an important role. Combining components will be a relatively easy task, enhancing efficiency in terms of man-hours needed to accomplish a given task. These pages, as well as the Metacomputer are still in a development stage …….. INFSO-RI-508833 Grid Tutorial, Groningen, September 2006 5 Example (3) Enabling Grids for E-sciencE Systems available at SARA in 1996 CRAY YMP Vector system IBM SP2 76 CPUs Parsytec CC 56 CPUs INFSO-RI-508833 Grid Tutorial, Groningen, September 2006 6 Example (4) Enabling Grids for E-sciencE • SARA news item on 16-6-1998 – Basis voor meta-omgeving gelegd. – Sinds 4 mei maakt SARA's IBM RS/6000 SP parallelle supercomputer gebruik van de DCE/DFS omgeving, een filesysteem dat een transparante computeromgeving mogelijk maakt. Met het nieuwe filesysteem zijn bestanden van DCE/DFS gebruikers wereldwijd toegankelijk met andere computersystemen die beschikken over DCE/DFS, waarmee een belangrijke basis is gelegd voor de meta-omgeving. – Gebruikers aan de VU science faculty hebben nu op een uniforme manier toegang tot hun bestanden, ongeacht of ze werken op de RS/6000 SP of een lokaal workstation. Hetzelfde geldt voor gebruikers van het Parsytec CC systeem bij SARA: vanaf zowel de Parsytec als de RS/6000 SP zijn alle bestanden voor de gebruiker direct toegankelijk. INFSO-RI-508833 Grid Tutorial, Groningen, September 2006 7 Example (5) Enabling Grids for E-sciencE • A web interface was developed for submitting jobs to the metacomputing environment, also a meta job language was used. • Also job migration between systems and mpi over two systems was investigated – First time we heard about globus, one of the well known building blocks now for grid infrastructures. – Network link between systems was a problem, only FE link, Gbit not available, HiPPI (800 Mbps) not available for Parsytec. INFSO-RI-508833 Grid Tutorial, Groningen, September 2006 8 Today (1) Enabling Grids for E-sciencE • So what is new today? – Scale! Grid infrastructures operate worldwide • International infrastructures - EGEE, DEISA, Nordugrid, OSG, TeraGrid • National – NAREGI (Japan), UK-eScience, D-Grid, NLGrid – Interoperability – availability of middleware – Globus toolkit, UNICORE, NAREGI, schedulers INFSO-RI-508833 Grid Tutorial, Groningen, September 2006 9 Today (2) Enabling Grids for E-sciencE • Some basic requirements for a grid infrastructure – Transparent user administration – single sign on (single grid identity), authorisation and accounting based on grid identity – AAA facilities – Job scheduling – which can handle different environments – Global data access – Global information services – job information, data information, resource information • Interoperability! – Standards needed for federation of infrastructures – GGF, IETF…. INFSO-RI-508833 Grid Tutorial, Groningen, September 2006 10 Networking (1) Enabling Grids for E-sciencE • Developments in network connectivity (high bandwidths) and tools play an important role – 10 Gbps WAN links available today, both shared links and dedicated lightpaths (based on lambda technology) – 1 Gbps network adapters are commodity items on systems today and 10GE adapters available INFSO-RI-508833 Grid Tutorial, Groningen, September 2006 11 Networking (2) Enabling Grids for E-sciencE – GridFTP can use multiple streams in order to take full advantage of available bandwidth – Parallel files systems can take full advantage of underlying high speed networks - throughput can be in the order of 100MByte/s and more – Tuning of WAN TCP must get attention, e.g. latencies are in the order of milliseconds (~20 in Europe), defaults on systems mostly not suited for bulk data transports. INFSO-RI-508833 Grid Tutorial, Groningen, September 2006 12 SURFnet 6 infrastructure Enabling Grids for E-sciencE SURFnet6 DWDM on dark fiber Groningen1 Middenmeer1 Den Helder Harlingen Leeuwarden Assen1 Dwingeloo1 Beilen1 Emmen1 IBG1 & IBG2 Emmeloord Subnetwork 4: Purple Lelystad2 BT Beilen1 Hoogeveen 1 Meppel 1 NLR Alkmaar1 DLO NLR Lelystad1 Haarlem1 Amsterdam1 Amsterdam2 Leiden1 B T Zwolle1 BT Muenster Subnetwork 3: Red Apeldoorn1 3XLSOP Subnetwork 1: Green DenHaag Breukelen1 Schiphol-Rijk Hilversum1 Arnhem Enschede1 Zutphen1 Utrecht1 Rotterdam4 Delft1 Wageningen1 Nijmegen1 Ede Rotterdam1 Dordrecht1 Breda1 Nieuwegein1 Den Bosch1 Eindhoven1 Maasbracht1 Tilburg1 Maastricht1 Geleen1 Heerlen2 Venlo1 Bergen-opZierikzee Zoom Middelburg Subnetwork 2: Dark blue Subnetwork 5: Grey Heerlen 1 Heerlen1 Vlissingen Krabbendijke INFSO-RI-508833 Grid Tutorial, Groningen, September 2006 13 Enabling Grids for E-sciencE NetherLight – Lightpath connections to the Netherlands GLORIAD-RU @NIKHEF GE 622M GLORIAD 3rd quarter 2005 INFSO-RI-508833 Grid Tutorial, Groningen, September 2006 14 Enabling Grids for E-sciencE Global Lambda Integrated Facility (GLIF) World Map www.glif.is INFSO-RI-508833 Visualization courtesy of Bob Patterson, NCSA/University of Illinois at Urbana-Champaign. Data compilation by Maxine Brown, University of Illinois at Chicago. Earth texture from NASA. Grid Tutorial, Groningen, September 2006 15 GEANT2 topology Enabling Grids for E-sciencE INFSO-RI-508833 Grid Tutorial, Groningen, September 2006 16 The EGEE project Enabling Grids for E-sciencE • EGEE – 1 April 2004 – 31 March 2006 – 71 partners in 27 countries, federated in regional Grids – Operation of a pan European production Grid • EGEE-II – 1 April 2006 – 31 March 2008 – Expanded consortium • 91 partners • 11 Joint Research Units – Natural continuation of EGEE – Emphasis on providing production-level infrastructure • increased support for applications • interoperation with other Grid infrastructures INFSO-RI-508833 • more involvement from Industry Grid Tutorial, Groningen, September 2006 17 The EGEE infrastructure Enabling Grids for E-sciencE • Mission – Manage and operate production e-Infrastructure open to all user communities and service providers – Contribute to Grid standardisation and policy efforts • Infrastructure operation – Currently include ~200 sites across 39 countries – Continuous monitoring of Grid services in a distributed global infrastructure – Automated site configuration/management • Future – Expand on interoperability with related infrastructures INFSO-RI-508833 Grid Tutorial, Groningen, September 2006 18 Enabling Grids for E-sciencE EGEE-II Federations and Countries INFSO-RI-508833 Grid Tutorial, Groningen, September 2006 19 Operational Organisation Enabling Grids for E-sciencE INFSO-RI-508833 Grid Tutorial, Groningen, September 2006 20 User Support Activities Enabling Grids for E-sciencE INFSO-RI-508833 Grid Tutorial, Groningen, September 2006 21 User support in NE region Enabling Grids for E-sciencE • NE website: http://www.egee-ne.org/operations • User support: contact user support at local site or mail to support@egee-ne.org – NE uses a ticketing system monitored by different partners from our region. In NL NIKHEF, RC-RuG, SARA responsible. – Tickets from GGUS are also imported in the NE system • Application support – NA4 activity. In NL RCRuG, SARA INFSO-RI-508833 Grid Tutorial, Groningen, September 2006 22 A Selection of Monitoring tools Enabling Grids for E-sciencE 1. GIIS Monitor 2. GIIS Monitor graphs 3. GOC Data Base 4. Scheduled Downtimes 5. GridIce – VO view 6. Live Job Monitor INFSO-RI-508833 Grid Tutorial, Groningen, September 2006 Enabling Grids for E-sciencE INFSO-RI-508833 Grid Tutorial, Groningen, September 2006 24 The DEISA project • Objective: To enable Europe’s terascale science by the integration of Europe’s most powerful supercomputing systems. • DEISA is an European Supercomputing Service built on top of existing national services. This service is based on the deployment and operation of a persistent, production quality, distributed supercomputing environment with continental scope. Grid Tutorial, Groningen, September 2006 25 THE DEISA SUPERCOMPUTING GRID AIX distributed super-cluster Vector systems (NEC, …) Linux systems (SGI, IBM, …) Grid Tutorial, Groningen, September 2006 26 Participating Sites BSC CINECA CSC ECMWF Barcelona Supercomputing Centre Consortio Interuniversitario per il Calcolo Automatico Finnish Information Technology Centre for Science European Centre for Medium-Range Weather Forecast Research Centre Juelich High Performance Computing Centre Stuttgart Institut du Développement et des Ressources en Informatique Scientifique - CNRS Leibniz Rechenzentrum Munich Rechenzentrum Garching of the Max Planck Society Dutch National High Performance Computing and Networking centre Spain Italy Finland UK UK (int) EPCC/HPCx University of Edinburgh and CCLRC FZJ HLRS IDRIS LRZ Germany Germany France Germany RZG SARA Germany The Netherlands Grid Tutorial, Groningen, September 2006 27 DEISA technologies • GPFS – parallel filesystem for transparent file access from all systems – dedicated European network used for high throughput • Loadleveler-MC for job submission on AIX systems • UNICORE for job submission to all systems • Common Programming Environment (CPE) on all systems for DEISA users • Single username on all systems Grid Tutorial, Groningen, September 2006 28 Access • Users can submit proposals for access to DEISA resources through DECI (DEISA Extreme Computing Initiative) calls • Proposals are evaluated by national committees and depending on ranking get access to resources • Most partners contribute about 10% of their resources for DEISA applications • URL: www.deisa.org Grid Tutorial, Groningen, September 2006 29