CUG SUMMIT 2002 Abstracts
SGI Focused Program

3A IRIX

SGI IRIX Update, Lynne Johnson, SGI

SGI has continued to provide quarterly releases for IRIX this year. This talk will cover our accomplishments in IRIX for the last year, and discuss our road map for the next year.

Experiences in Managing System Resources on a Large Origin 3800 Cluster, Mark van de Sanden and Huub Stoffers (SARA)
Paper in PDF format
Slides in PDF format

SARA has been running a 1024 CPU Origin 3800 cluster since early 2001. During that year the system underwent many changes. We gained a lot of experience in running a system of this size. In this talk we will present an overview of our experiences in managing and controlling resources like cpu, memory, system buffers in order to optimize the system's performance. The following items will be discussed as illustration: cpusets, jlimits, nbuf, LSF, disk space, PVM, MPI.

How Should the IRIX Cpuset Feature Be Employed on a 256P Origin3800? Barry Sharp and James Glidewell (BCS), and Greg Siekas, SGI
Paper in PDF format
Slides in PDF format

This paper examines and explores the use and benefits of the IRIX Cpuset feature on a 256P Origin3800 system. The question of how to use the Cpuset feature is raised as it was unclear just how well the feature inter-operated with batch subsystems such as NQE, PBS Pro and LSF. Our particular interest was with Veridian System's PBS Pro so the focus in this paper is on this rather than the other two. Some perceived deficiencies in the Cpuset feature are discussed and enhancements to the feature are offered as suggestions.

3B Applications

Development of Vadose Zone DNAPL Transport Models, Eric Greenwade (INEEL)

The movement of dense, non-aqueous phase liquids (DNAPLs) in the unsaturated zone is of great interest throughout the US and the rest of the world. These fluids are generally left over from various industrial processes and have traditionally been disposed directly into the ground. The movement of these liquids in the subsurface above the water table is extremely complex and difficult to apply on the field scales necessary to make informed decisions. This talk will discuss a new research initiative focused on scaling up molecular dynamics type calculations for inclusion in field-scale models. Existing models are also evaluated as to their appropriateness.

The Parallel Performance of a Tightly Coupled 3D Magnetohydrodynamic Simulation, Lee Margetts and Mike Pettipher (MCC)
Paper in PDF format
Slides in PDF format

In magnetohydrodynamics two distinct processes, magnetism and fluid flow, are physically tightly coupled. Traditional methods start by decoupling these processes, solving the fluid and electromagnetic problems separately. The physical coupling is achieved later using the results of the fluid calculation in the electromagnetic problem and vice versa. We depart from this approach, solving the full system of equations using a tightly coupled parallel program. We compare the performance of this program on various architectures, in particular the Cray T3E and Origin3000.

An Overview of Recent Supercomputing Applications at JPL, Chris Catherasoo (JPL)

The Jet Propulsion Laboratory (JPL) is responsible for much of NASA's unmanned space exploration activities including astronomy, astrophysics, earth observations and planetary explorations. The Supercomputing and Visualization Facility at JPL provides the various projects with a state-of-the-art supercomputing and visualization environment for modeling and data analysis. Our facilities include a 16-processor Cray SV1-1A, an SGI Origin2000 (with 128 processors and 6 graphics pipes), a StorageTek tape silo and a Powerwall room with seven projectors. Examples of recent work that has been done on the JPL supercomputers include Martian rover navigation simulations, Shuttle Radar Topography Mission (SRTM) data analysis, ocean modeling and spacecraft engine and heat-shield simulations. This presentation will cover some of these recent activities.

3C Visualization

SGI High Performance Visualization Solutions, Fabrizio Maguglian, SGI

The increase in computational power is generating more and more results and data, and leads to the requirement to visually analyze huge volumes of data in an interactive and user-friendly manner. The SGI Onyx 3000 and Onyx 300 series of scalable systems are focused on making such a task possible. The SGI Onyx 3000 now offers a choice of industry leading graphics subsystems. InfiniteReality3 offers the highest image quality available, and InfinitePerformance graphics offers the highest polygon/second and non-anti-aliased pixel/second performance available. Recently announced, SGI Onyx 300 delivers InfiniteReality3 graphics at more affordable price points and denser rack configurations than ever before.

OpenGL Vizserver™, George Melissargos, SGI

OpenGL Vizserver is an application transparent solution for visual serving and collaborative visualization. The presentation explores when, where and how Vizserver is used as the key component in implementing Visual Area Networking. The newly released 2.0 version is described in detail, together with the main issues concerning its deployment and the new features that will appear in the future.

Wavelets and Their Application to Very Large Data Visualization, John Clyne (NCAR)

This talk will provide an introduction to wavelets and explore their use for multi-resolution data representations suitable for interactive visualization of very large volume data.

4A Linux

SN-McKinley Project Update, Steve Reinhardt, SGI

SGI is developing parallel supercomputers based on its innovative NUMAflex interconnect, with the significant innovations of Itanium Processor Family (IPF) processors from Intel and the Linux operating system. This talk will describe progress on the project, things we've learned about these new technologies, and future directions for the project.

Resource Management on a Mixed Processor Linux Cluster, Haibo Wang (MCSR)
Paper in PDF format
Slides in PDF format

We built a mixed processor Linux cluster with 16 Pentium III and 41 Pentium IV processors.To maximize the performance and utilization of the system, we implement a schedule according to job type and resource requirement. The resource issues related to computation chemistry software such as NWCHEM, and MPQC are reported in this paper.

4B Performance

1024p Single System Image Origin 3000, Chuck Niggley (NAS)

NASA Ames and SGI have worked together over the last 6 years building successively larger CC-NUMA systems. This talk will discuss progress with the latest system in development, a 1024 processor single system image machine. Topics will include lessons learned, scaling, performance, and the current operational status of the machine.

High Performance Linear Algebra, Adrian Tate (MCC) and Patrick Briddon, University of Newcastle upon Tyne
Paper in PDF format
Slides in PDF format

Scalapack is the standard parallel linear algebra library, though we will give an example of its use directly inhibiting performance and scalability. We will demonstrate the replacement of a particular Scalapack routine which simplifies the linear algebra involved in the tridiagonalisation of matrices via 1-sided communications. In addition to the simplification of the linear algebra involved, the natural performance gain of such methods has a direct influence upon the scalability of the code in question.

Obtaining Scalable Performance from Molecular Dynamics Codes on HPC Machines, Neil Stringfellow, Peter Coveney, and Fabrizio Giordanetto (MCC)

A large amount of computational time allocated on high performance computers is used for molecular dynamics simulations, and there is increasing demand for scalable codes particularly in the field of large molecule simulations in disciplines such as biochemistry. This paper investigates the performance of the latest generation of molecular dynamics codes on high performance machines and discusses design strategies which have led to increased scalability. Examples are given which demonstrate the ability of these codes to solve problems which scale in size with the number of processors available, as well as their scalability in terms of decreasing execution time.

4C Mass Storage

SGI CXFS Clustered File-System and Its Place in Storage Management, Neil Bannister, SGI

Silicon Graphics released its Clustered File-System, CXFS, at the end of 1999. Since that date it has gained good acceptance among SGI customers and it continues to be deployed for large storage requirements. This year SGI is in the process of rolling out CXFS on other platforms. This paper will give an overview of CXFS and how it is being used by customers. It will also cover how it is integrated into SGI's storage solutions, current status and future plans.

Parallel I/O Experiences on an SGI 750 Cluster, Troy Baer (OSC)

This paper will describe OSC's experiences in using parallel I/O for scientific applications on a cluster of SGI 750 Itanium systems connected with Myrinet and Gigabit Ethernet. The target file systems include local disks on each node, NFS served by an SGI Origin 2000, and PVFS served by 16 Pentium III systems. The parallel I/O applications demonstrated include several standard I/O benchmarks (bonnie, ROMIO perf, NAS btio, and ASCI Flash) as well as user application codes.

SGI's Complex Data Management Architectures, LaNet Merrill, SGI

This presentation will help ensure you get the ultimate performance and reliability between all your storage hardware and software. It will help you choose and integrate technologies like RAID, Fabrics, CXFS, DMF, NFS and FailSafe for SAN, NAS or DAS architectures

5A-C Tutorials

Resource Management Using Cpusets, Sam Watters, SGI

The cpuset system provides for predictable run-times by allowing CPU and memory resources to be allocated for specific users or processes. When used in conjunction with a workload management system, the result is a flexible space-sharing solution. This presentation will discuss the IRIX cpuset system, the cpuset API, issues of concern when using cpusets, and future directions for cpuset support in IRIX. The new capabilities concerning memory migration as processes are moved between cpusets and when checkpointed processes are restarted will be described in detail.

MPI Performance Tuning, Karl Feind, SGI

This talk is geared toward the MPI user interested in getting best performance from the SGI MPI library on IRIX. An overview will be given of the optimizations that the MPI library does automatically for the user. Some optimizations are not enabled by default, and these will be presented along with tips on when and how to use them.

Optimization Techniques in SGI Scientific Libraries, Mimi Cellis, John Baron, and Arthur Raefsky, SGI

SCSL, SGI's Scientific Computing Software Library, is a collection of optimized and parallelized mathematical functions and procedures commonly used in high performance computing, including basic linear algebra operations, dense matrix solvers, eigenvalue/vector routines, signal processing routines (such as fast Fourier transforms), and direct and iterative sparse solvers. Although SCSL provides a very specific set of functionality, the tuning techniques employed are of general applicability for HPC codes. In this talk we discuss some of the optimization methods used in SCSL, including hand-unrolling loops to expose computations, cache-blocking for matrix operations, FFTs, and sparse solvers, and preventing false-sharing in parallel codes.

6A Operating Systems and Operations

Transitioning from NQE to PBS Pro, James Glidewell and Barry Sharp (BCS)

Appendix to this paper

Slides in PDF format

This paper presents what was involved in transitioning from using NQE to PBS Pro on an Origin3800 system. Conversion aspects such as user impacts, new scheduling policies, use of cpusets, running NQE and PBS Pro in parallel, differences and working with a new vendor Veridian are also described.

Experiences with LSF and Cpusets on the Origin3800 at the Dresden University of Technology, Stefanie Maletti (DRESDENU)

Based on the experience in a long ownership of a Cray T3E and an Origin2800 and their usage for parallel applications, an operational concept was designed for the Origin3800 at the Dresden University of Technology. The exclusive usage of processors by means of LSF 4.1 and cpusets is the main item of this concept. We present our problems starting with the installation and configuration of LSF 4.1 with cpusets till the enforcement of our operational concept.

8->64->256 in 275 days!, Barry Sharp and James B. Glidewell (BCS)

This paper presents experiences of upgrading an SGI Origin system from 8 to 64 and then 256 processors and 4 gigabytes to 384 gigabytes of memory within a nine-month period. These experiences cover such things as working with SGI to reach agreement on unadvertised hardware partitioning schemes that they were willing to support, evaluating the use of CXFS across partitions, CXFS and DMF limitations and IRIX issues dealing with large memory sizes. The highlight of successfully joining a 192P upgrade to an existing 64P system without severe interruption to production is also described.

NCSA'S World Class Machine Room, Brian Kucic (UIUCNCSA)

An inside look through the design and construction process of NCSA's new 7500 sq.ft. machine room. This machine room has capacity for today's high performance processors while providing expansion for future needs.

6B Programming Environments

MIPSPro Compiler Update, Jim Galarowicz, SGI

MIPSpro 7.3.1.3m was released in December 2001 and is the third maintenance update to the MIPSpro 7.3 compilers. This presentation provides an overview of the content of this release as well as an introduction into next generation MIPSpro 7.4 compilers.

Message Passing Software Status and Plans, Karl Feind, SGI

SGI message-passing software has been enhanced in the past year in the areas of performance optimizations, additional MPI-2 support, and support for new communication interconnect drivers. This presentation describes the recent enhancements to message-passing and related software which includes MPI, SHMEM, PVM, and Array Services. Future planned features are also described.

SGI Developer Tools Software Status and Plans, Jim Galarowicz, SGI

SGI developer tools continue to be enhanced. This presentation will describe the recent enhancements to the debugger and performance tools. The talk will include information about the current WorkShop, SpeedShop, and dbx releases. An overview of the planned enhancements for next release of SGI developer tools will also be included.

Etnus TotalView Debugger, Mary Kay Bunde, Etnus

TotalView has been used on Cray and SGI machines alike for years, though TotalView has undergone surprising changes over the last year. Topics covered in this session include: a bit of history, for those confused about Cray TotalView vs. Etnus TotalView, TotalView 5.0 features, a preview of TotalView 5.1, and hints of long term direction.

6C User Services

Worldwide Collaborative Research and Learning Environments: The Access Grid Experience, Leslie Southern (OSC) and Mary Bea Walker (UIUCNCSA)

This paper will introduce the Access Grid (AG), a suite of technologies designed to support high performance scientific collaboration. The authors will show how this technology is used to provide training on high performance computing topics to a diverse community. The authors will discuss how they developed a series of Web-based tutorials that enable technical users to join the AG worldwide community. The AG was created by researchers at Argonne National Laboratory, USA, and is in continuous development by members of the AG community.

Supporting Users at NASA Ames Research Center, Chuck Niggley, John Pandori, and Chris Hamilton (NAS)
Paper in PDF format
Slides in PDF format

The NASA Advanced Supercomputing (NAS) Division User Services Group provides front line support and administrative help to more than 2000 active users and staff. We manage 28 SGI Origin supercomputer systems, as well as a recently arrived Cray SV1ex. There are, as well, more than 25 file and/or Web servers, 6 tape servers managing more than 100 tape devices and 25 RAID file systems. In order to accomplish our mission, we make use of any and all tools that the various vendors provide, as well as developing many specialized monitoring tools of our own. We have systems in place that set up & maintain user accounts across all of our systems, manage the scheduling of maintenance or special activities on individual systems and keep users informed of system status change in a timely manner. This paper will describe the operations, tools and procedures that we use in accomplishing all of the above tasks and many others that haven't been mentioned.

Code Optimization and Parallelization on the Origins–Looking From Users' Perspective, Yan-Tyng Sherry Chang (NAS)
Paper in PDF format
Slides i nPDF format

While the pioneers are working hard to get the best performance out of the large Origins for large applications, most of the general users are still in the dark or are struggling to learn even the basic techniques to optimize/parallelize their codes. In this paper, some experiences will be presented on learning the optimization/parallelization techniques from SGI documentations. Emphasis of this paper will be on a few essential issues (with examples) that general users should master when they work with the Origins.

OpenMP and MPI on Multiple Platforms, Terry Nelson (NAS)
Paper in PDF format
Slides in PDF format

This paper discusses the differences of running OpenMP applications on the Cray SV1, C90, and SGI Origins. Experiences with the conversion of user codes to OpenMP will be presented. MPI and OpenMP coexistence will also be discussed.

7A SIG Open Meeting

Operating Systems, Barry Sharp (BCS)

This session will be of interest to those who want the inside scoop on SGI's plans for current and future Irix releases. SGI representatives will discuss their development plans for IRIX, LINUX, DMF, CXFS, and support for LSF and PBS, etc. These questions and more will be answered in a thumbnail sketch of development plans for Irix and Security. Following this 15 minute presentation, we will discuss features we both like and dislike about the existing products. These will be categorized, and the top likes and dislikes for each area will be presented to SGI. This meeting will also be an opportunity to express to SGI what kinds of future conference presentations we would like to have from them.

Introductions of Focus Group Chairs

SGI thumbnail sketch of future plans
Open discussion of issues and concerns; wish list discussion
Voting on the top issues and concerns to present to SGI

7B SIG Open Meeting

Programming Environments, Hans-Hermann Frese (ZIB)

The Programming Environments SIG invites you to attend its Open Meeting on Tuesday afternoon. After a brief introduction of the SIG's business and Focus areas, we shall start off with a lively discussion on various issues concerning Programming Environments, Compilers and Libraries, and Software Tools. Attendees will have the opportunity to discuss their concerns with the liaisons and technical experts from SGI. Your feedback for the SIG's business and recommendations on future activities will be gratefully received.

7C SIG Open Meetings

Computer Services/User Services, Leslie Southern (OSC)

The Computer Services SIG is open to all CUG participants who have interests in the daily operations and support of computing centers. The session includes vendor updates and is primarily devoted to open discussions about issues related to such topics as services, training, consulting, documentation, outreach, support, and operations. After a brief introduction of SIG and Focus Chairs, and a short business meeting, participants will have an opportunity to describe their computing environment and support requirements as well as target areas for potential improvement.

Computer Services/Operations, Leslie Southern (OSC)

The Computer Services SIG is open to all CUG participants who have interests in the daily operations and support of computing centers. The session includes vendor updates and is primarily devoted to open discussions about issues related to such topics as services, training, consulting, documentation, outreach, support, and operations. After a brief introduction of SIG and Focus Chairs, and a short business meeting, participants will have an opportunity to describe their computing environment and support requirements as well as target areas for potential improvement.

8A Interactive Sessions

IRIX Security, Casey Schaufler, SGI
Paper in PDF format
Slides in PDF format

The Irix Operating System provides a number of advanced security facilities which are unsurpassed in the marketplace. Many of these facilities have been used to provide unique solutions in the supercomputer environment. The Trusted Irix/CMW (Trix) feature can be used to provide extreme levels of resource sharing with data integrity and protection.

IRIX Security Wish List, Virginia Bedford and Liam Forbes (ARSC)

One goal of the Security Focus Group is to generate a prioritized list of security related action items that the CUG membership would like the vendors to work on. The Security Focus Group Chairs and SGI are hosting a discussion to develop the first iteration of the IRIX Security Wish List. Bring your list of IRIX security issues (RFEs, bugs, or insights), and let SGI know what you would like to see in the future of IRIX security.

8B Interactive Session

Visualization Theatre, John Clyne (NCAR)

The CUG Visualization Theatre is an opportunity to showcase your center's best efforts, and learn about the work of others, through short, informal data visualization animations. The Visualization Theatre is open to all CUG attendees. Contributors are reminded to please contact the Visualization Focus Chair, John Clyne (clyne@ncar.ucar.edu) by Monday, May 6th