GRID ENGINE WORKSHOP

April 22 - 24, 2002

Sun Microsystems Gridware GmbH

Dr.-Leo-Ritter-Straße 7 D-93049 Regensburg, Germany

CONTENTS
 

Participants

Agenda

Day 1 - Monday, April 22

09:00	Welcome (Fritz Ferstl)
09:15	Sun Grid Computing Strategy (Wolfgang Gentzsch)
10:15	Grid Engine, Overview and Roadmap (Fritz Ferstl)
11:00	Break
11:30	The Grid Engine Project (Fritz Ferstl)
12:00	Lunch
13:15	Working in the Grid Engine Project - Part I (André Alefeld)
14:45	Break
15:15	Working in the Grid Engine Project - Part II (Andreas Haas)
17:00	Adjourn
19:00	Dinner Invitation ("Leerer Beutel" - Bertold Straße - see map)

Day 2 - Tuesday, April 23

09:00	Partner Grid Engine Projects
	Integration and Support for Grid Engine Martin Kloock, Cards Engineering  Workstation Cluster at Ford of Europe Craig Stair, Raytheon Integration, Support and Development Projects Related to Grid Engine
	Grid Computing (Globus etc.) & Grid Engine
	Geoff Cawood, EPCC Using Sun Grid Engine and Globus to Schedule Across a Combination of Local and Remote Machines Krzysztof Kurowski, PSNC  "Gridlab" and "Progress" as the Examples of Cooperation Between SGE and Grid Research at PSNC
10:30	Break

11:00	Partner Grid Engine Projects cont'd
	Grid Engine and Portals Kenneth Edgecombe, HPCVL Secure Grid and Portal Computing Eric Stahlberg, OSC Sun TCP Bioinformatics Application Integration on Heterogeneous Server Platforms Steven Newhouse, Imperial College  Portals and Resource Scheduling at Imperial College P. M. Dew, Univ. of Leeds The White Rose Computational Grid

12:30	Lunch
13:30	Partner Grid Engine Projects cont'd
	Scheduler Enhancements for Grid Engine and other Topics Andrea Lorenz, RWTH Aachen Scheduling in an HPC Environment Soren-Aksel Sorensen, UCL-CS Covering the Spectrum - UCL-CS Grid Related Activities Babu Sundaram, University of Houston Overview of set up of the Campus Grid, EZ-Grid system at UH
14:30	Adjourn
16:00	Walk through Regensburg
17:30	At people's disposal

Day 3 -  Wednesday, April 24

09:00        Project Wrap Up; Build Special Interest Groups

09:30        Special Interest Group Meetings

11:00        Break

11:30        Special Interest Group Meetings cont'd

12:30        Lunch

14:00        Presentation of SIGs

15:00        Roundtable discussion

15:45        Closing Remarks (Fritz Ferstl)

16:00        End
 
 
 

Presentations

Sun Grid Computing Strategy and Projects

Sun Grid Computing Strategy

By Wolfgang Gentzsch, Sun Microsystems Inc.

The Grid Engine Project

By Fritz Ferstl, Sun Microsystems Gridware

Working in the Grid Engine Project: Part I

By André Alefeld, Sun Microsystems Gridware

Working in the Grid Engine Project: Part II

By Andreas Haas, Sun Microsystems Gridware

Partner Grid Engine Projects

Integration and Support for Grid Engine

Workstation-Cluster at Ford of Europe

By Martin Kloock, Cards Engineering

Abstract:

In August 1998 cards Engineering was able to place an on-site support contract at Ford Motor Company in Cologne/Merkenich. I started there maintaining the CAE (Computer Aided Engineering) application server. About 50 CAE applications are installed on their server which is a SUN Ultra Enterprise 450. My activity covered the installation of the applications, 1st- and 2nd-level helpdesk and the license maintenance of all these products. Additionally I had to collaborate very closely with the Ford colleagues in Dunton/UK and Dearborn/USA.

During this time some Ford people thought about a better usage of their workstations (w/s), mainly HPs and altogether about 800 w/s. Additionally they thought they could decrease the costs the Cray supercomputer was producing.

About at the same time SUN acquired Gridware and we saw a good opportunity to help Ford with introducing a resource management system. With our close connections to SUN we were able to introduce SGE at Ford and to create a workstation cluster in two development departments at Ford. After one of my colleagues from the Cologne cards PIT team took over my job maintaining the CAE server I was able to work on the w/s cluster based on a new contract.

In these months from summer 2001 until April 2002 SGE was installed on about 80 w/s and adapted 8 CAE applications (i.e. Radioss, Nastran, Fluent) for the workstation cluster for using them in serial (1 CPU or w/s) and/or parallel mode. This had to be done by changing startup scripts, GUI configurations and creating submit scripts. Additionally I configured the cluster due to user's wishes.

With the help of some SUN guys we were able to convince the responsible persons at Ford of Europe that SGE is a good product which does exactly what they requested. Unfortunately we were not able to convince the Ford of US people completely. Over there LSF and PBS, two competitive products, are very strong and already used before we began with SGE. Nevertheless it was possible to install a productive w/s cluster in Cologne and Dunton/UK and to satisfy the users concerning their needs.

Integration, Support and Development Projects Related to Grid Engine

By Craig Stair, Raytheon

Abstract:

Raytheon’s High Performance Computing(HPC) team has been engaged in large-scale supercomputer-based systems development, deployment, operation, and maintenance for 30 years. The team works continuously to maintain its comprehensive knowledge of IBM, SGI, SUN, Compaq, HP, Cray/NEC, and end-to-end storage solutions/file systems. Raytheon’s in-house technical strengths for high-end systems development includes systems architecture engineering, system performance engineering, modeling, applications performance engineering, facilities engineering and implementation.

Grid Computing (Globus etc.) & Grid Engine

Using Sun Grid Engine and Globus to Schedule Across a Combination of Local and Remote Machines

By Geoff Cawood and Paul Graham, Edinburgh Parallel Computing Centre, EPCC

Abstract:

The aim of this collaboration between EPCC and Sun is to produce a job scheduler (based on SGE) which can submit jobs to both local and remote machines. Globus will provide a secure means of running jobs on remote sites.
The project is still at an early fact-finding stage so this talk will compare some potential solution strategies and invite comments from the floor on the technical issues arising.

GridLab and Progress as the Examples of Cooperation between SGE and Grid Research at PSNC

By Krzysztof Kurowski, Poznan Supercomputing and Networking Center, PSNC

Abstract:

Grid research and development at Poznan Supercomputing and Networking Center (PSNC) will be presented during the SGE Workshop. Amongst different projects the presentation will be focused on GridLab and Progress projects and their relationships to SGE.
Two important aspects of Grid technology, which have been largely ignored, form the basis of the GridLab Project, which aims to build components for Grid applications, and realistic testbeds for their development: Co-development of Infrastructure and Applications and Dynamic Grid Computing.
In case of Progress project, scientific portal and Grid environment for SUN Servers will be developed based on SUN technologies. The main goal of this project is to provide reliable and easy access to the Grid as well as support research community with many useful tools and services.

Grid Engine and Portals

Secure Grid and Portal Computing

By Kenneth Edgecombe, HPCVL

Abstract:

HPCVL was formed by a consortium of four universities (Carleton University, Queen's University, the Royal Military College of Canada, and the University of Ottawa) to provide a secure innovative High Performance Computing (HPC) environment for researchers. This environment is now being built with contributions from the four member institutions, the Canada Foundation for Innovation, the Ontario Innovation Trust, the Ontario Research and Development Fund, Sun Microsystems, IBM, and Entrust.
The progress being made, the roadmap for future developments, and the acceptance by researchers will be reviewed. The partnership with Sun Microsystems will be highlighted and progress on the Centre for Secure Grid and Portal Computing discussed.

Sun TCP Bioinformatics Application Integration on Heterogenous Server Platforms

By Eric Stahlberg, Ohio Supercomputer Center, OSC

Abstract:

A brief overview of efforts made to integrate applications of running on a variety of server architectures without using SGE will be presented. Proof-of-concept integration approaches for using the Technical Compute Portal with Time Logic DeCypher bioinformatics algorithm accelerators, Cray and SGI origin systems will be discussed. Issues and potential problems associated with the implementations will be highlighted.
 

Portals and Resource Scheduling at Imperial College

By Steven Newhouse, Imperial College, London

Abstract:

The London e-Science Centre at Imperial College, one of the eight Regional Centres within the UK e-Science program, is involved in the construction of Grid infrastructures within the College and in the UK through the use of SGE, Globus and its own grid middleware - ICENI.
As part of the Centre's activities, and as a Sun Centre of Excellence, we have been developing middleware to support an e-Science Portal at the College (based on uPortal), the use of the Technical Compute Portal to access Sun Grid Engine through the uPortal infrastructure, and the integration of scheduling infrastructures (such as SGE) through a web service enabled DRMAA interface.
The presentation will comprise an overview of this work and its future directions.

The White Rose Computational Grid

By P. M. Dew, University of Leeds

Abstract:

This talk will describe the manner in which the Sun Grid Engine (SGE) and the Technical Computing Portal (TCP) are to be used in support of collaborative e-Science projects at the White Rose universities. Our presentation will include an example of the UK e-Science project, DAME (Distributed Aircraft Maintenance Environment), which aims to deliver a grid test-bed for distributed diagnostics. The project builds on the White Rose Computational Grid, based on the Sun Grid Engine, and the Technical Computing Portal that will provide a Web-based interface to technical applications executed on any of the White Rose Computational Grid nodes.

Scheduler Enhancements for Grid Engine and other Topics

Scheduling in an HPC environment

By Andrea Lorenz, RWTH Aachen

Abstract:

With the purchase of RWTH's SunFire equipment the old queueing system (GNQS) was replaced by SGE.
The operational part works right out of the box, however, for our job load characteristics with a broad range of job sizes with respect to memory and required CPUs, the standard scheduler gives only a poor utilization.
In the talk the specific requirements and our approaches to meet them are presented.

Covering the Spectrum Grid Activities@UCL

By Søren-Aksel Sørensen, University College of London, UCL-CS

Overview of Sun Center of Excellence in Geosciences at the University of Houston

By Babu Sunderam, University of Houston

Abstract:

University of Houston (UH) is a Sun Center of Excellence in Geo sciences. Recently, University of Houston obtained a hardware grant and setting up a campus grid to aggregate resources across the campus. Major participants include the HPCTools research group lead by Dr. Barbara Chapman, High performance computing center, Advanced Geosciences lab and Mechanical engineering. At this workshop, we will present an overview of the hardware and software setup of this campus grid under construction. Also, we explain the EZ-Grid system being developed at UH for easier setup, managing and efficient usage of grid environments. The major components of the system include integrated resource broker, usage policy management frameworks and information services. The interfaces with Globus toolkit and Sun Grid Engine are to be presented. We also expect to receive feedback from the workshop participants on improving our EZ-Grid system and tightly integrating with Sun Grid Engine software.

Special Interest Group Meetings

Globus Integration and Brokers

• Objectives/plans/achievements of the organizations participating in the SIG.
• User requirements; expected scenarios to be solved.
• Action items.

• EPCC: Integrate Globus and Grid Engine with broker on top. Project has started. Prototype to be available by end of 2002. Project to be finalized in 2 years.
• Imperial College: Integrate Grid Engine with Globus via web services and Jini. Brokerage needed for own operation of site – better sooner than later.
• Univ. of Houston: Build broker for Globus and integrate with Grid Engine. Gram enhancements. Prototype already exists.
• PSNC: Grid architecture based on Globus integrating Grid Engine and self developed broker. Special requirement is that Grid service requirements can appear within already executing jobs. Project is in architecture definition phase.
• Raytheon: Has concrete customers which are looking for Grid architectures and solutions today.
• Sun: Looking for the best technology to take to the market and to solve the problems of existing and future customers with Grid requirements.

• Setup a SIG mailing list (F. Ferstl).
• Each participant to submit requirements, use cases and objectives as available.
• Each participant to submit designs, concept papers, etc. as available.

TCP Interest Group

• Current State & what's missing
• Goals
• Short term goals

• iPlanet Portal server is currently required
• installation currently sometimes difficult, at least not straightforward, needs cleanup
• installation under root -> root access should be singled out in a special controlled component, the rest should for security reasons not run as root
• adding of new applications needs a lot of admin interaction and adoptions, so it is currently not easy
• MAC browser cannot work with TCP (netlets)
• NT support not available
• configurability not too easy
• missing documentation, buglist, FAQ
• shared filesystem needed in the current implementation
• Unix account necessary
• file organization of TCP generated files not structured enough
• hook to an existing security framework/environment missing
• prototype is running quite stable and reliable
• setup changes are requiring some amount of admin work, should be easier
• access of password protected web content from a channel was not possible
• Javascript is required
• navigation within forms should be enhanced (too many windows popping up/down)
• customizable views for different users dependent on their authentication/authorization record should be possible
• workflow integration would be nice, macro functionality to replay a set of actions
• netlet technology available, but has restrictions (with VNC e.g. OpenGL not working); should be available to get visualization from a host where the data are or which has the corresponding capabilities

• utility functions should be supported by the portal (e.g. converters, visualizers etc.)
• better output handling should be available (e.g. visualization from any machine through VNC or whatever, output file handling should be more sophisticated to suppress unnecessary file transfers)
• independence from iPlanet
• On what standards is TCP based and who is defining them ?
• ease of installation/administration/application integration
• user profiles should exist and be connected to a security infrastructure, logging and trace facilities should be available
• user space customization and application adding should in certain limits be possible
• better job monitoring facilities should be available
• web service support
• EJB support for persistence

• create TCP (shall be renamed) Open Source Project with mailing list, source code repository (project will be hosted on GE Open Source site)
• around end of May/beginning of June the current version of TCP with some smaller changes will be available (if the legal issues concerning the O'Reilly Servlet Book code have been clarified)
• project guidelines (java programming guidelines, I18N support, etc.) shall be made available
• Legal issues shall be clarified

Parallel Jobs and Scheduling

What concepts does SGE provide for parallel scheduling?

Sun Grid Engine 5.3, Administration and User's Guide (Chapter: Managing Parallel Environments, see page 271: The Allocation rule input..) and sge_pe (5)

Sun Grid Engine 5.3, Administration and User's Guide (Chapter: Tight Integration of PEs and Sun Grid Engine Software, page 277)

None

Scheduler Documentation

Closing Discussion

• Grid visions for the mid and long term future
• Applications suitable for Grids