TUTORIAL TUTORIAL Rebecca Breu, Bastian Demuth, Andr Giesler, - PowerPoint PPT Presentation

TUTORIAL TUTORIAL Rebecca Breu, Bastian Demuth, André Giesler, Bastian Tweddell (FZ Jülich) {r.breu, b.demuth, a.giesler, b.tweddell}@fz-juelich.de September 2006

Outline Outline � Motivation & History � Production UNICORE 5 � Features � Recent Developments � Architecture � UNICORE 6 � Grid Standards � Architecture � Summary 2

Motivation: Why UNICORE? Motivation: Why UNICORE? � Scientists need computational and storage related resources 3

Motivation: Why UNICORE? Motivation: Why UNICORE? � Supercomputers are managed by Resource Management Systems (RMSs) that handle the scheduling � But: There are many RMSs available � A proprietary way of job submitting Resource Management System � IBM Loadleveler � llsubmit … � Torque Resource Manager � qsub … � Different job description languages (# of nodes, memory requirements…) 4

Motivation: Why UNICORE? Motivation: Why UNICORE? � Solution: Grid System UNICORE � Define job workflows in abstract manner � Immediate portability of job definitions for other systems with other architectures � No ‘learn overhead’ if a new RMS is used � Applications across multiple supercomputers/clusters � ‘going meta’ 5

Initial UNICORE developments developments Initial UNICORE � UNICORE 08/1997-12/1999 � UNICORE Plus 01/2000-12/2002 � EUROGRID 11/2000-01/2004 � GRIP 01/2002-02/2004 � OpenMolGRID 09/2002-02/2005 6

UNICORE in (mostly) European Projects UNICORE in (mostly) European Projects Chemomentum eDEISA A-WARE OMII-Europe EGEE-II DGI UniGrids DEISA VIOLA OpenMolGRID GRIDSTART GRIP 2002 2003 2004 2005 2006 2007 2008 2009 7

UNICORE Features UNICORE Features 8

Recent Developments Recent Developments � OGSA-based and WSRF-compliant UNICORE 6 alpha � Interactive access � UNICORE-SSH plugin � High-level API for programming Grids � DRMAA-based access to RMSs � Comfortable configuration tool � UNIX-like command line interface 9

UNICORE Live CD UNICORE Live CD � Complete “out-of-the-box” usage of UNICORE 5 � Bootable Linux OS with UNICORE 5 pre-installed � Does not harm your system � sandbox scenario � For testing, evaluating, … AVAILABLE HERE AND AS ISO-IMAGE ON SOURCEFORGE 10

UNICORE Architecture Overview UNICORE Architecture Overview Client SSL Multi-Site Jobs opt. Firewall opt. Firewall Gateway Gateway Authentication opt. Firewall Abstract NJS NJS NJS Authorization Authorization UUDB UUDB IDB IDB Incarnation IDB Non- TSI TSI TSI Abstract RMS Disc RMS Disc RMS Disc Vsite Vsite Vsite Usite Usite 11

UNICORE Architecture Overview UNICORE Architecture Overview � Workflow-Engine Client � Resource Management � Job-Monitoring SSL Multi-Site Jobs � File Transfer opt. Firewall opt. Firewall � User Management Gateway Gateway Authentication � Application Support opt. Firewall Abstract NJS NJS NJS Authorization Authorization UUDB UUDB IDB IDB Incarnation IDB Non- TSI TSI TSI Abstract RMS Disc RMS Disc RMS Disc Vsite Vsite Vsite Usite Usite 12

UNICORE Architecture Overview UNICORE Architecture Overview Client SSL Multi-Site Jobs opt. Firewall opt. Firewall Gateway Gateway Authentication opt. Firewall Abstract NJS NJS NJS Authorization Authorization UUDB UUDB IDB IDB Incarnation IDB Non- TSI TSI TSI � similar to Globus jobmanager Abstract � fork � LoadLeveler, (Open)PBS(Pro), RMS Disc RMS Disc RMS Disc CCS, LSF, NQE/NQS, ... Vsite Vsite Vsite � CONDOR, GT 2.4 Usite Usite 13

UNICORE Architecture Overview UNICORE Architecture Overview Client SSL Multi-Site Jobs opt. Firewall opt. Firewall Gateway Gateway Authentication opt. Firewall Abstract NJS NJS NJS Authorization Authorization UUDB UUDB IDB IDB Incarnation IDB � similar to /etc/grid-security/grid-mapfile Non- TSI TSI TSI Abstract RMS Disc RMS Disc RMS Disc Vsite Vsite Vsite Usite Usite 14

UNICORE Architecture Overview UNICORE Architecture Overview Client SSL Multi-Site Jobs opt. Firewall opt. Firewall Gateway Gateway Authentication opt. Firewall Abstract NJS NJS NJS Authorization Authorization UUDB UUDB IDB IDB Incarnation IDB Non- TSI TSI TSI Abstract RMS Disc RMS Disc RMS Disc Vsite Vsite Vsite Usite Usite 15

Management Workflow Vsites Usites UNICORE Client UNICORE Client 16 Monitoring Preparation Job Job

UNICORE Architecture: Client UNICORE Architecture: Client UNICORE Client � Definition of abstract Jobs (not machine-specific) Scientific � Creation of complex workflows (e.g.multi-site jobs for UNICORE CLI Applications client tier Abstract Jobs geographically dispersed supercomputers) AJO and UPL over SSL Usite A Usite C optional firewall Gateway authentication server tier Multi-site Jobs Vsite Y Vsite X UUDB Network authorization Job and Supervisor incarnation IDB server tier Non-abstract Jobs Target submit System and / or Interface execution scheduling Resource and Management resource System management target system tier Vsite Z Usite B 17

UNICORE Architecture: Client UNICORE Architecture: Client UNICORE Client � Client extensions via application-specific plugins Scientific � Various plugins have been developed over the last UNICORE CLI Applications client tier Abstract Jobs years: AJO and UPL over SSL Usite A Usite C � Car-Parrinello Molecular Dynamics Plugin optional firewall (speed up quantum chemical computations) Gateway authentication � Fluent, Gaussian, MSC Nastran Plugins server tier Multi-site Jobs Vsite Y � OpenMolGrid Plugins (Molecular Science) Vsite X UUDB Network authorization � PovRay Plugin Job and Supervisor incarnation IDB server tier Non-abstract Jobs Target submit System and / or Interface execution scheduling Resource and Management resource System management target system tier Vsite Z Usite B 18

UNICORE Architecture: Client UNICORE Architecture: Client UNICORE Client Scientific UNICORE CLI Applications client tier Abstract Jobs AJO and UPL over SSL Usite A Usite C optional firewall Gateway authentication server tier Multi-site Jobs Vsite Y Vsite X UUDB Network authorization Job and Supervisor incarnation IDB server tier Non-abstract Jobs Target submit System and / or Interface execution � Based upon the Arcon Client API (Java-API) scheduling Resource and Management resource System management target system tier Vsite Z Usite B 19

UNICORE Architecture: Gateway UNICORE Architecture: Gateway � Gateway UNICORE Client Scientific � Single entry-point for UNICORE Sites (Usites) UNICORE CLI Applications client tier � Contains n Virtual Sites (Vsites) Abstract Jobs AJO and (e.g. provides access to n supercomputers) UPL over SSL Usite A Usite C � Only ONE OPEN PORT in firewall optional firewall Gateway authentication � Authentication of users via X.509 certificates server tier Multi-site Jobs � Perfect base to group resources and access Vsite Y Vsite X within Virtual Organizations (VOs) UUDB Network authorization Job and Supervisor IDB incarnation server tier G Non-abstract Jobs A T Target submit System and / or E Interface execution W scheduling Resource and A Management resource System management Y Virtual Organization target system tier Vsite Z Usite B 20

UNICORE Architecture: NJS UNICORE Architecture: NJS UNICORE Client � Network Job Supervisor (NJS) Scientific � Analyse defined Workflows (workflow engine) UNICORE CLI Applications client tier Abstract Jobs � Multi-site, multi-step Jobs AJO and UPL over SSL Usite A Usite C � Directed Acyclic Graphs optional firewall Gateway authentication server tier � Submit of sub-jobs to other sites over Gateway Multi-site Jobs Vsite Y Vsite X UUDB Network authorization � Job Incarnation via Incarnation Database (IDB) Job and Supervisor incarnation IDB server tier � translate abstract-jobs into machine-specific jobs Non-abstract Jobs � E.g. location of preinstalled software packages Target submit System and / or Interface execution scheduling Resource � Authorization via Unicore User Database (UUDB) and Management resource System management target system tier � turn X.509 user certs to users xlogin of machine Vsite Z Usite B � The complete certificate is checked (not only DN ) 21

UNICORE Architecture: TSI UNICORE Architecture: TSI UNICORE Client � Target System Interface Scientific � Represents a real Target System UNICORE CLI Applications client tier Abstract Jobs (e.g. a supercomputer or storage server) AJO and UPL over SSL Usite A Usite C optional firewall � Interaction with Resource Management Systems Gateway authentication server tier � Torque, PBS Pro, Load Leveler, (UNIX fork), … Multi-site Jobs Vsite Y � Job submit to Resource Management System Vsite X UUDB Network � Scheduling via Resource Management System authorization Job and Supervisor incarnation IDB � Retrieve job results and job output server tier Non-abstract Jobs � Manage workspace Target submit System and / or Interface execution scheduling Resource and Management resource System management target system tier Vsite Z Usite B 22