2^nd GRID ENGINE WORKSHOP

September 22-24, 2003

For a second time, the SGE Development Workshop brought together Grid Engine users, developers and development partners. They came from sixteen organizations, out of research and industry, in eight different countries. The attendees participated in introductory sessions to current and future Grid Engine developments. They had the opportunity to give an overview of their own Grid Engine projects, and to learn about others' Grid interests, areas of focus and results. Breakout sessions on several topics further stimulated discussion of common interest. Additional Grid Engine Development Workshops will be proposed on the project mail lists.

Participants

Afeldt, Stefan MSC Software GmbH Am Moosfeld 13 D-81829 München Germany Tel: +49 (0)89 431987 161 Fax: +49 (0)89 431987 7 161 stefan.afeldt@mscsoftware.com

Alefeld, André Sun Microsystems GmbH Dr. -Leo-Ritter-Straße 7 D-93049 Regensburg Germany Tel: +49 (0)941 3075 255 Fax: +49 (0)941 3075 222 andre.alefeld@sun.com

Bablick, Ernst Sun Microsystems GmbH Dr. -Leo-Ritter-Straße 7 D-93049 Regensburg Germany Tel: +49 (0)941 3075 135 Fax: +49 (0)941 3075 222 ernst.bablick@sun.com

Beltrami, Riccardo Chiron S.r.l. Via Fiorentina, 1 I-53100 Siena Italy Tel: +39 0577 243388 riccardo_beltrami@chiron.it

Berger, Martin Sun Microsystems GmbH Dr. -Leo-Ritter-Straße 7 D-93049 Regensburg Germany Tel: +49 (0)941 3075 225 Fax: +49 (0)941 3075 22 martin.berger@sun.com

Bhardwaj, Aastha Sun Microsystems Inc. 7777 Gateway Blvd, Bldg 12 Mailstop: NWK12-310 Newark, CA 94560 USA Fax: 001 510 936 3571 aastha.bhardwaj@sun.com

Cambruzzi, Sandro Sun Microsystems GesmbH Wienerbergstrasse 3 A-1100 Wien Austria Tel: +43 (1)60563 11903 Fax: +43 (1)60653 11920 sandro.cambruzzi@sun.com

Cawood, Geoff University of Edinburgh EPCC James Clerk Maxwell Building Mayfield Road Edinburgh EH9 3JZ UK Tel: +44 (0)131 650 5818 Fax: +44 (0)131 650 6555 g.cawood@epcc.ed.ac.uk

Dalton, Terry National Research Council of Canada Institute for Marine Bioscienses 1411 Oxford Street Halifax, Nova Scotia Canada Tel: +1 902 426-9995 Fax: +1 902 426-9413 terry.dalton@nrc-cnrc.gc.ca

Danesi, Silvio Sun Microsystems S.p.A. V. Le Fulvio Testi, 327 I-20162 Milano Italy Tel: +39 02 64152 592 silvio.danesi@sun.com

Davidson, Shannon Raytheon P.O. Box 660023 Dallas, Texas, 75266-0023 USA svdavidson@swbell.net

Dörr, Andreas Sun Microsystems GmbH Dr. -Leo-Ritter-Straße 7 D-93049 Regensburg Germany Tel: +49 (0)941 3075 123 Fax: +49 (0)941 3075 222 andreas.doerr@sun.com

Edgecombe, Dr. Kenneth Queen's University High Performance Computing Virtual Laboratory HPCVL Queen's University 141 Collingwood Street Kingston, ON K7L 3X6 Canada edgecomk@post.queensu.ca

Ferstl, Fritz Sun Microsystems GmbH Dr. -Leo-Ritter-Straße 7 D-93049 Regensburg Germany Tel: +49 (0)941 3075 110 Fax: +49 (0)941 3075 222 fritz.ferstl@sun.com

Furmento, Dr. Nathalie London E-Science Centre Imperial College Department of Computing 180 Queen's Gate London SW7 2BZ UK Tel: +44 (0)207-594-8310 Fax:+44 (0)207-581-8024 n.furmento@doc.ic.ac.uk

Gabler, Joachim Sun Microsystems GmbH Dr. -Leo-Ritter-Straße 7 D-93049 Regensburg Germany Tel: +49 (0)941 3075 233 Fax: +49 (0)941 3075 222 joachim.gabler@sun.com

Gentzsch, Dr. Wolfgang Sun Microsystems Inc. 18 Network Circle Menlo Park, CA 94025 USA Tel: +1 650 786 2032 wolfgang.gentzsch@sun.com

Grell, Stephan Sun Microsystems GmbH Dr. -Leo-Ritter-Straße 7 D-93049 Regensburg Germany Tel: +49 (0)941 3075 226 Fax: +49 (0)941 3075 222 stephan.grell@sun.com

Haas, Andreas Sun Microsystems GmbH Dr. -Leo-Ritter-Straße 7 D-93049 Regensburg Germany Tel: +49 (0)941 3075 131 Fax: +49 (0)941 3075 222 andreas.haas@sun.com

Hardie, Duncan Sun Microsystems Ltd. Medomsley Road Consett, DH8 6TJ UK Tel: +44 (0)1207 585 504 Fax: +44 (0)1207 591 002 duncan.hardie@sun.com

Januszewski, Radoslaw Poznan Supercomputing and Networking Center ul. Noskowskiego 10 Pl-61-704 Poznan Poland Tel: +48 61 858 2053 Fax: +48 61 852 5954 radoslaw.januszewski@man.poznan.pl

Kosiedowski, Michal Poznan Supercomputing and Networking Center ul. Noskowskiego 10 Pl-61-704 Poznan Poland Tel: +48 61 858 2161 Fax: +48 61 852 5954 michal.kosiedowski@man.poznan.pl

Lees, Peter Sun Microsystems Australia Pty Ltd. 828 Pacific Highway Gordon, New South Wales Australia Tel: +61 2 9844 5375 Fax: +61 2 9418 2014 peter.lees@sun.com

Lobodzinski, Bogdan DESY Zeuthen Platanenallee 6 D-15738 Zeuthen Germany Tel: +49 33762 7 7359 Fax: +49 33762 7 7216 bogdan.lobodzinski@desy.de

Lorenz Andrea Aachen University of Technology Center for Computing an Communication Seffenter Weg 23 D-52074 Aachen Germany Tel: +49 (0)241 80-29791 Fax: +49 (0)241 80-22504 lorenz@rz.rwth-aachen.de

Markov, Lev Sun Microsystems Inc. 1, Network Drive Building #2 Burlington, MA 01803 USA Tel: +1 781 442 8547 Fax: +1 781 442 1692 lev.markov@sun.com

McBride, David London E-Science Centre Imperial College Department of Computing 180 Queen's Gate London SW7 2BZ UK Tel: +44 (0)207-594-0 Fax:+44 (0)207-581-8024 dwm99@doc.ic.ac.uk

Meier, Dr. Ulrich Sun Microsystems GmbH Brandenburgerstr. 2 D-40880 Ratingen Germany Tel: +49 2102 4511593 Fax: +49 2102 499516 ulrich.meier@sun.com

Mikolajczak, Rafal Poznan Supercomputing and Networking Center ul. Noskowskiego 10 Pl-61-704 Poznan Poland Tel: +48 61 858 2053 Fax: +48 61 852 5954 rafal.mikolajczak@man.poznan.pl

Nelson-Gal, David Sun Microsystems Inc. 4170 Network Circle Santa Clara, CA 95054 USA Tel.: +1 650 352-8520 david.nelson-gal@sun.com

Newhouse, Steven Imperial College Department of Computing 180 Queen's Gate London SW7 2BZ UK Tel:  +44 (0)20 7594 8316 Fax: +44 (0)20 7581 8024 s.newhouse@doc.ic.ac.uk

Piwowarek, Pawel Poznan Supercomputing and Networking Center ul. Noskowskiego 10 Pl-61-704 Poznan Poland Tel: +48 61 858 2036 Fax: +48 61 852 5954 pawel.piwowarek@man.poznan.pl

Raghunath, Priya University of Houston Department of Computer Science 501, PGH 4800 Calhoun Rd Houston Texas 77204 USA Fax: 001 713 743 3335 praghuna@cs.uic.edu

Ryan, Paul Sun Microsystems Ltd. Regis House 45 King William Street London, EC4R 9AS UK Tel: +44 161 9058 137 paul.d.ryan@sun.com

Schwierskott, Andy Sun Microsystems GmbH Dr. -Leo-Ritter-Straße 7 D-93049 Regensburg Germany Tel: +49 (0)941 3075 200 Fax: +49 (0)941 3075 222 Andy.Schwierskott@sun.com

Seed, Thomas University of Edinburgh EPCC James Clerk Maxwell Building Mayfield Road Edinburgh EH9 3JZ UK Tel: +44 (0)131 650 6485 Fax: +44 (0)131 650 6555 t.seed@epcc.ed.ac.uk

Shirin, Gregory Sun Microsystems Inc. 18 Network Circle Menlo Park, CA 94025 USA Tel: +1 650 786 6907 gregory.shirin@sun.com

Sørensen, Dr. Søren-Aksel University College of London Department of Computer Science Gower Street London WC1E 6BT UK Tel: +44 (0)171 380 7285 Fax: +44 (0)171 387 1397 S.Soerensen@cs.ucl.ac.uk

Street, Stefano University College of London Department of Computer Science Gower Street London WC1E 6BT UK Tel: +44 (0)20 76793701 Fax: +44 (0)20 73871397 s.street@cs.ucl.ac.uk

Templeton, Daniel Sun Microsystems GmbH Dr. -Leo-Ritter-Straße 7 D-93049 Regensburg Germany Tel: +49 (0)941 3075 220 Fax: +49 (0)941 3075 222 dan.templeton@sun.com

Tierney, Dr. Craig High Performance Technologies, Inc. 11955 Freedom Drive Reston, VA 20190 USA Tel: +1 703 707 2700 ctierney@hpti.com

Turner, Aaron University of York Dept. of Computer Science York, YO1 4 BE UK Tel: +44 (0) 1904 432763 aaron@cs.york.ac.uk

Youhanaie, Fredrick Oxford Supercomputing Centre Oxford University Parks Road Oxford OX1 3 QD UK Tel: +44 (0) 1865 283569 Fax. +44 (0) 1865 273839 Fred.Youhanaie@comlab.ox.ac.uk

Agenda

DAY 1 - Monday, September 22 - Sorat Hotel

Grid Engine 6.0 Development

DAY 2 - Tuesday, September 23 - Sorat Hotel

Grid Engine Partner Developments and Applications

DAY 3 - Wednesday, September 24 - Sun Microsystems Regensburg Offices

Special Interest Group Day

Presentations

Grid Engine 6.0 Development

Grid Engine Project Report and Workshop Outline

    By Fritz Ferstl, Sun Microsystems GmbH

Multi-Threading Approaches in Grid Engine 6.0

    By Andreas Dörr, Sun Microsystems GmbH

Grid Engine 6.0 Cluster Queues

    By Ernst Bablick & Andreas Haas, Sun Microsystems GmbH

Grid Engine 6.0 Throughput Scheduler and Scheduling Improvements

    By Stephan Grell, Sun Microsystems GmbH

Utilizing Database in Grid Engine 6.0

    By Joachim Gabler, Sun Microsystems GmbH

Sun Grid Engine 6.0 Accounting / Reporting Tool

    By André Alefeld, Sun Microsystems GmbH

Status of the Grid Engine DRMAA Implementation

Status of the Grid Engine DRMAA Implementation (Java Binding)

    By Andreas Haas and Dan Templeton, Sun Microsystems GmbH

Reservation / Preemption / Backfilling in Grid Engine 6.0

    By Andreas Haas, Sun Microsystems GmbH

Graph Optimization Algorithms for Grid Engine

    By Lev Markov, Sun Microsystems Inc.

Resource Discovery in Sun Grid Engine using JXTA

    By Dan Templeton, Sun Microsystems GmbH

Impact of Sun Grid Solution Selling Initiatives

    By Glenn Wright, Sun Microsystems Inc.

High Availability Grid with SGE and Sun Cluster

    By Sandro Cambruzzi, Sun Microsystems GesmbH

Plenary Discussion: Feedback and Comments for SGE 6.x

Grid Engine Partner Developments and Applications

Challenges in building OGSA based Grids

    By Steven Newhouse, Imperial College London

Abstract:

Key to wide-scale adoption of the Grid is to provide an infrastructure which can be used to build viable business models. The UK's Markets for Computational Services (MCS) project is working to build such an infrastructure using the Open Grid Services Architecture.
The presentation reports on our recent implementation experiences in producing the first prototype using OGSI.

Using Resources of Multiple Grids with Grid Service Provider

    By Michal Kosiedowski, Poznan Supercomputing and Networking Center (PSNC)

Abstract:

Recently, there have emerged numerous grid installations and systems around the globe. Most of them are equipped with a graphical user interface, preferably a web based one, to submit and execute computing jobs. Yet, users of these multiple grids must switch from one computing portal to another to utilize resources that are available to them. It requires a big effort to migrate a job description from one grid to another.
Poznan Supercomputing and Networking Center proposes a new mechanism to enable resources of multiple grids within a user interface. The PROGRESS grid service provider, which provides grid services sharable between distributed computing portals and other grid user interfaces, is set to be equipped with a grid resource broker plug-in solution. The grid resource broker plug-ins will provide mechanisms for communication with a particular grid system, e.g. PROGRESS grid resource broker, Globus, GridLab, etc. They will serve as gateways to the grids whenever there's a requirement to submit a computing job to the grid and enable job descriptions migration. In this presentation the overall functionality of the PROGRESS grid service provider is delivered and the idea of the grid resource broker plug-in mechanism is introduced.

EPCC Sun Data and Compute Grids

    By Geoff Cawood, Edinburgh Parallel Computing Centre

Abstract:

Authors: Geoff Cawood, Ratna Abrol, Thomas Seed, Terry Sloan

The SunDCG project aims to develop a compute and data scheduler based around Grid Engine, Globus and a variety of data technologies.
Last year's Grid Engine Workshop provided helpful feedback on some proposed technical strategies. This year, our presentation will describe the resulting software, and our plans for a further release when the project ends in January 2004. We hope to share some of our experiences of integrating Grid Engine and Globus, and seek any technical insights that could ease the final development phase.

Access to Shared Grid Resources in Heterogeneous Queuing Systems

    By Pawel Piwowarek, Poznan Supercomputing and Networking Center (PSNC)

Abstract:

Authors: Pawel Piwowarek, Marek Zawadzki

PSNC is a leading provider of computational power in Poland. To deal with extensive HPC demands, all requests are served via heterogeneous queuing systems such as LSF, LL, PBS, etc. which help sharing resources among users. With the advent of PROGRESS project in 2002 PSNC gained another grid-like infrastructure which utilize SGE.
This talk presents our solution of integrating SGE deployed in PROGRESS into our production environment based on the LSF platform.

Update on Sun CoE Activities at the University of Houston

    By Priya Raghunath, University of Houston

Abstract:

As a Sun Center of Excellence in Geo sciences, University of Houston (UH) has been actively pursuing research and development in a grid environment. This presentation will focus on the development of the EZ-Grid system at the UH that provides a generic interface to access grid resources. Having undergone design changes the new version of EZ-Grid is implemented using the latest technologies and provides an enhanced set of features. We will also present a brief overview of the job manager we have developed to enable interaction between Globus and SGE. In addition to these we will briefly describe efforts underway to develop strategies for increased fault-tolerance without compromising on performance and also proposed tools to enable prediction of average wait times for jobs.

The White Rose Grid: Experiences and “Wish List”

    By Aaron Turner, University of York

Abstract:

A brief outline of the experience at the York White Rose Grid node, with an outline of the set up used, and some problems encountered. Based on this experience a series of things that it would be good to see in release 6.0 are discussed.

Integrating SGE and Globus in a heterogeneous HPC environment

    By David McBride, Imperial College London

Abstract:

This talk describes the integration of Sun Grid Engine and the Globus Toolkit on the London e-Science Centre's computational resources.
It presents an overview of the the architectural implementation details of these systems and also presents some of the challenges encountered when trying to deploy our solution on production systems.

ICENI: A next generation grid middleware

    By Nathalie Furmento, Imperial College London

Abstract:

This talk presents how ICENI, the Grid Middleware developed in the London e-Science Centre, and SUN products such as uPortal, SGE or Netbeans, have been integrated to provide solution to applied scientists in delivering them access to Grid computing resources. It also presents the implementations of the ICENI's Service-Oriented Architecture and the Semantic Framework attached to it.

Integration of SGE into NPACI Rocks

    By Najib Ninaba, Scalable Systems Pte Ltd., SCS Linux Competence Center

(Presented by Andy Schwierskott, Sun Microsystems GmbH)

Experiences at MSC Software in Applying Sun Grid Engine in a CAE Environment

    By Stefan Afeldt and Stefan Mayer, MSC Software

Abstract:

During the past years, workload management has become a very important topic in the CAE (Computer Aided Engineering) environment. In June 2003 Sun announced MSC.Software's plans to market, implement and support Sun Grid Engine and to offer associated services to help its manufacturing customers worldwide reduce the time and costs associated with product development. In our presentation, we describe MSC.Software's recent work around Sun Grid Engine. For example, MSC.Software has implemented an interface to Sun Grid Engine in its MSC.BatchSubmit offering. This tool offers a web-based interface between users of CAE software and Grid Engine, so that engineers can easily access CAE applications in a compute farm from any web browser on any workstation platform, independent of the workstation architecture. In addition, MSC.Software developed Grid Engine administration modules for the popular Webmin environment, so that system administrators can easily set up and configure a Grid Engine installation from a web browser Based on this development work, MSC.Software has implemented Sun Grid Engine based solutions both in multiple customer service projects and at MSC. Software's European headquarters in Munich, where 200 MSC engineers have access to a compute farm of more than 100 CPUs.

Aspects of Processing a Grid

    By Søren Sørensen, University College London

Scheduling and Job Management using Grid Engine on a Multi-Teraflop HPC

    By Craig Tierney, High Performance Technology Inc.

Confessions of a Grid Engine Addict

    By Shannon Davidson, Raytheon

Abstract:

Grid Engine Enterprise Edition is used at several HPC sites managed by Raytheon. An overview at how GEEE is used at these sites will be presented. A Myrinet / MPICH GEEE integration is presented and a Grid demo for SC2003 is described.

Sun Grid Engine at OSC

    By Fred Youhanaie, Oxford Supercomputing Centre

Abstract:

Sun Grid Engine was first installed at Oxford Supercomputing Centre in March, 2002. Since then the installation has evolved from the default configuration to one that handles multiple clusters with multiple Parallel Environments, with the ability to accept jobs submitted through the Globus middleware. This talk will chart our experiences with the Grid Engine software over the past 18 months.

Interactive Batch Jobs and AFS Token Handling Mechanism

    By Bogdan Lobodzinski, DESY Zeuthen

Abstract:

After an introduction about the usage of the SGEEE batch system at DESY Zeuthen we present the realization of interactive jobs managed by SGEEE on hosts without direct login access. In addition the AFS token handling mechanisms is covered: the current implementation, further developments and problems are discussed.

Moving to a Secure Grid Portal

    By Kenneth Edgecombe, High Performance Computing Virtual Laboratory, Kingston, Can.

Special Interest Group Day

DRMMA SIG

Discussion points were:

What will be the advantages of a DRMMA standard?

• General facilitation of Grid adoption 

• Particular facilitation for ISVs

• 1 API for many DRMs

• Offer of self contained solutions

What will / could be helpful to get the standard generally accepted?

• DRM adoption

• ISV adoption

• Ease of use

• Eventually a certification (ISV, DRM)

Who are the DRMAA contributors?

• SGE

• PBS

• IBM Loadleveler

• United Devices

• Entropia

• Cadence (ISV)

• GridIron

What is the particular interest of the SIG participants in DRMMA?

• Grid Solutions

• Generic integration to different types of functions (Condor, LSF, SGE, ...) 

• Finding alpha testers for SGE implementation

• Promotion of standards within Sun 

• Promotion of choices

Grid Engine Portal SIG

The integration of GEP into a Security Infrastructure like Entrust have been discussed and this is a project underway at HPCVL.
The problem here are the different layers where security is deployed by the different components of the whole system, which is rather complicated and only some aspects could be discussed.

The second topic that had been covered was the need for easier addition of applications and how to describe them in a generic way.

One possible solution would the definition of an application definition language and a corresponding tool to allow the integration of any application into a portal and to automatically generate the needed configuration dialogues.

Three kinds of users would be useful.

• Users that administrate GEP and add new applications and define which users can access it.

• Special expert users that are doing tuning for special applications and find out which parameter sets are best suited, this can be a longer running effort.

• Users that are only deploying the preconfigured application delivering only a minimal set of input and data.

OGSA / OGSI SIG

Discussion points were:

Definition and need of OGSA / OGSI?

• What are OGSA/OGSI?

• Why are they useful?

• Should SGE support OGSA?

• Should SGE support OGSA?

Objectives / Plans / Achievements

OGSA/OGSI are nice, but functionality is more useful than standards.

If SGE had remote monitoring/controlling features, most people would be happy.

XML output from SGE command line utils would be a huge step in the right direction.

An OGSI interface would be more useful than OGSA support.

OGSI binding for DRMAA a start, but need broader functionality.

Requirements/Scenarios

There are two basic needs:

• Hierarchical grid site structure

• Two-tier grid site structure with central meta-scheduler

It is possible to do both without OGSA/OGSI

Action Items

• Look into XML output for command line utilities

• Work towards OGSI binding for DRMAA Goals

• Enable meta-scheduling in SGE

Scheduler SIG

The meeting was divided into three parts:

Current used configurations of the different grids and job characteristics:

• Usually some large, more medium, and most short running jobs

• Short running: 10 to 15 min (usually not parallel jobs)

• Long running : over 8 hours ( usually massive parallel jobs)

• Each combined with different sets of resource requests and dead lines

• In some environments have jobs a max runtime of 24 hours. They have to restart themselves, if they need longer. This is used to prevent starvation of jobs.

• Most grids are build with SGE and not with SGEEE.

Problems with the current SGE feature set and different work arounds:

General feedback was positive. Most have some additional scripts to handle specific requirements and things SGE cannot do (such as: a pre scheduler, ...)

Some wishes for enhancements were:

• Better license management for parallel jobs with consumables

• A way to limit the number of jobs per host.

• A better way to specify job dependencies

Advanced reservation and back filling, explained by Lev Markov:

A short summery of Lev Markov's and Andreas Haas' presentation with the focus on job starvation, resources and back filling.

Closing Remarks

Dear Conference Participants, Dear Colleagues,

I want to thank everybody who participated and contributed to our Second International Workshop on Grid Engine Technologies, September 22 -24, 2003, in Regensburg, and made it a real success.

With 50 participants, 30 experts presentations, and several Workshops on the last day, the Conference was certainly packed, interesting, and very interactive. Besides presentations about new Grid Engine core technology, there was also a great collection of work around Grid Engine, e.g. Globus, OGSA, Data Grids, Cluster, Campus and Global Grids, Advanced Scheduling, Security, and more.

I would like to especially thank our esteemed technology partners who provided a lot of important input to Grid Engine, thus helping us to develop an advanced and competitive next-generation DRM technology.

Final, special thanks to Monika Grobecker and Fritz Ferstl, who again organized the Workshop and contributed to a very successful event.

I hope to meet you all at the next Regensburg Workshop :-)

Kind Regards

Wolfgang Gentzsch
Director Grid Computing, Sun Microsystem