CUG SUMMIT 2002 Abstracts
Cray Focused Program

3A Operating Systems

The SV2's Operating System was derived from several Cray operating systems. This presentation will discuss what the SV2 Operating System is, and what it is not. We will discuss the SV2's OS heritage, and possible evolutionary paths into the future. There will be an overview of resiliency, scalability, resource management and other Data Center Functionality requirements.

This presentation will describe the scheduling concepts of the SV2 operating system. These concepts present a system view similar to the familiar T3E while preserving the flexibility of a system with a global address space that is physically distributed. The presentation will cover aspects of the operating system as well as aspects of the political scheduling demon.

Cray Inc. has put together a clustering offering using Intel hardware with an enhanced software stack and is offering the cluster with Professional Services for integrating, installation, hardware and software maintenance. Cray believes that it has numerous differentiators in the offering. This talk will cover all aspects of the offering.

3B Solutions (Applications)

A scalable parallel Synthetic Aperture Radar (SAR) software correlator has been implemented on the T3E-900 at the Arctic Region Supercomputing Center. The implementation produces a full scene (100 km2) image in 90 seconds realizing a 26x speedup using only 28 processors. This algorithm was included in an interferometric processing system capable of creating interferograms in little more than 15 minutes. The presentation will discuss implementation details of this system focusing on the latency-hiding and parallel techniques that allowed an overall 12x speedup for what initially seemed a stubbornly serial code. Plans for extension of the system to include digital elevation model creation, deformation mapping capabilities, and near real-time volcano hazard monitoring will also be presented.

Currently multiple Grid systems are developed to satisfy one need: provide access to distributed computer resources. The GRIP (Grid Interoperability Project) approach is to combine the strengths of two different infrastructures, namely UNICORE and Globus, to give users comfortable access to a wide range of Grid systems.

The widely-used Monte Carlo particle transport code MCNP has been ported to the MTA-2 with good performance. The large, flat shared memory of the MTA-2 offers unique advantages for large scale simulations of this type. Performance of MCNP on a range of platforms will be presented and discussed.

4A Interactive Session

For more than 25 years Cray has been at the forefront of high-performance computing, contributing substantially to the advancement of knowledge and the quality of human life. Cray has designed, built, and supported high performance computing solutions for customers around the world. Cray Professional Services can help ensure the success of supercomputer implementations by partnering with customers to provide complete solutions for their challenging scientific and engineering problems. These solutions exploit Cray's deep supercomputing expertise and sterling reputation for quality. Cray HPC Solutions can provide software development, custom hardware, extensions to Cray supercomputing products, integration of third-party products, and skilled manufacturing and assembly. Cray is open to new ideas and computing challenges that most firms will not even attempt. Cray's experienced staff is ready to handle the most complex tasks related to innovation, reliability, and a commitment to quality.

The Cray SV1ex memory system, which is the latest enhancement to the Cray SV1 product line, began shipping to customers in November, 2001. This was the second major enhancement to the SV1 product line since initial deliveries began in June of 1999. This interactive session will focus on new features, capabilities and performance of the SV1ex. Additionally, Cray will discuss our Bioinformatics initiative which relies heavily on the Cray unique hardware features of the SV1 architecture.

4B Interactive Sessions

This will be an open, interactive session during which the status of the SV2 will be discussed. Other topics to be discussed will include migration from other Cray platforms, SV2 administration and tools, and other SV2 topics of interest to the audience.

We are nearing completion of initial software for the Cray SV2. To help prioritize the next stage of development, Cray intends to survey some of its current customers. The purpose of this interactive session is to discuss the survey results in detail, ensuring Cray has the right perspective on software requirements going forward.

5A Operating Systems

This paper discusses the VPN network configuration issues, installing a Boeing modified UNICOS OS, managing, operating and using a remote T916 in Chippewa Falls while waiting for delivery of a T932. As the network spanned the Internet, special security precautions had to be made and are discussed. Part of the attraction with the network configuration was that the remote T916 appeared as just another node on Boeing's intranet making it extremely easy for Boeing engineering people to use. The issues involved with operating and controlling the remote T916 from Seattle presented some challenges and are discussed in the paper. The experience of working with CRAY to successfully use their T916 in this manner is also presented. A short description of the T932 with two IOS-Es and 8 Gwords of SSD along with what CRAY had to do to make all 8 Gwords of SSD usable as SDS is given. A performance comparison of a T90 vs. NEC's SX-5/6 for a Boeing CFD code is provided as the CRAY T932 and NEC SX-6 were both considered for this HPC upgrade.

This paper reviews the open source security tools used here at ARSC to manage our Unicos, Unicos/mk, and IRIX systems. Our intention is to make managing differing platforms as similar as possible and to add security features not available in a standard OS distribution. Using Kerberos, SSH, Sudo, Swatch, Tcpwrappers, and Tiger, we've created a strong overall security profile and simplified managing the security on more than 70 systems. The modifications we've made so that the tools will work together securely, or even compile in some cases, are discussed within the framework of our local software management practices.

After replacing a Cray T90 with an SV1ex system at FZJ we observed variations in the CPU time of user applications that were unexpectedly high. The paper describes details of our observations, gives some information on the reasons for the performance variations and outlines measures taken to reduce the observed effects. Components involved in this process include the SV1ex memory hardware, system software and also user application codes.

5B Programming Environments

UPC is an emerging set of extensions to the C programming language for parallel programming. Cray will be supporting UPC extensions on the SV2. This talk will briefly describe the extensions that will be supported in the initial SV2 delivery and how these extensions can be used to improve parallel programming productivity and performance.

The SV2 architecture offers powerful processors running at several levels of parallelism. This talk will describe the parallel programming models available in the initial delivery of the SV2 and the techniques for deriving the best performance from SV2 processors through its compilers.

This paper will contrast autotasking and OpenMP as programming models, comparing the tools available, the ease of code development and the final performance obtained. The platform considered will be the SV1ex but experiences of portability with others such as the SGI3800 will be included.

5C Visualization

The CrossGrid project will develop, implement and exploit new Grid components for interactive compute and data intensive applications like simulation and visualisation for surgical procedures, flooding crisis team decision support systems, distributed data analysis in high-energy physics, air pollution combined with weather forecasting. The elaborated methodology, generic application architecture, programming environment, and new Grid services will be validated and tested thoroughly on the CrossGrid testbed, with an emphasis on a user friendly environment. The work will be done in close collaboration with the Grid Forum and the DataGrid project to profit from their results and experience, and to obtain full interoperability. The primary objective of this Project is to further extend the Grid environment to a new category of applications of great practical importance, and into 11 new European countries.

6A Mass Storage

There are several fast new technologies competing for network traffic. Regardless of which one is selected, the implied transaction rates for hosts rapidly reach levels that are difficult to sustain with traditional protocol implementations. The solutions to this problem involve various forms of protocol “offloading” that will be examined from an HPC perspective. Data storage devices, networks, and data storage protocols are also proliferating. They variously provide large capacities, large transaction rates, large single-stream bandwidth, large aggregate bandwidth, multiple host connectivity, data sharing, low latency, data security and recovery, fast transparent data backup, or other advantages. Each of these requires its own form of software support. The question to be considered is which of these initiatives should be supported with scarce software development resources to best support HPC customers.

This talk will first describe in some detail the SV2 I/O architecture, including network and mass storage options and functionality. Next, implementation progress and initial performance measurements will be discussed. Finally we will explore plans for future development as well as some alternative directions that might be taken as the SV2 product matures.

We are in process of moving our DMF storage on our Cray SV-1 from STK Redwood to 9940 drives/media. We thought a site talk on our experience/tips might be of interest. This would cover the OS configuration and the DMF custom scripts. Also problems and tips as we complete this task. We are also moving half of our SGI Origin 2000 DMF storage from 9840 drives/media to 9940 drives/media and might have a few tips there as well.

This paper will review the efforts of OSC and Cray as tests were run against different Sun hardware and software in an effort to determine some configuration options for an HPC infrastructure to support Cray systems. Issues with SANs, Networking and Storage management will be reviewed by the project team and discussed in this presentation. NOTE: The project is beginning in January, with results available by the CUG Meeting. The actual content of the paper may change based on testing.

6B Performance

The flow simulation program URANUS has been developed to calculate non-equilibrium flows around space vehicles reentering the earth's atmosphere. The program simulates not only the supersonic flow but also the chemical reactions occurring during reentry as they have a non-negligible influence on the flow. As a result of the calculation the heat flow and the heat load at the surface is of main interest. The performance of the parallel multi-block program is compared on a wide range of high performance architectures like CRAY T3E, HITACHI SR8000 and NEC SX5.

I will explain the issues involved in optimizing an existing code Phrap, to run on the Cray MTA. Phrap assembles shotgun DNA sequence data, making extensive use of the Smith Waterman algorithm to find the optimal alignment of pairs of strings. Its performance will be compared to other architectures.

High Performance Fortran permits to write parallel programs with much less programming effort than standard communication libraries such as MPI or PVM. Relative performance of compiled HPF programs is considered low, though. We show that a compiled HPF application will gain a substantial run-time improvement, if compilation incorporates properties of a specific hardware architecture into the final program. Our prototype HPF compiler “KarHPFn” inserts communication primitives of the Cray T3E into the target programs. Those programs run up to 30 times faster than their counterparts compiled with Portland Group HPF.

At ARSC, the upgrade path from a J90 to SV1ex included installation of an SV1 followed by the first SV1e processor upgrade at a user site and then the first SV1ex memory upgrade. I will present results from performance analysis of user codes throughout this upgrade-in-stages and show that the SV1ex increase in memory bandwidth, and consequent improvement of machine balance, has helped more ARSC users than the CPU upgrade. I'll describe code attributes, such as memory and cache efficiency, which proved useful in predicting the response of individual applications to the hardware upgrades.

7A SIG Open Meeting

This session will be of interest to those who want the inside scoop on CRAY's plans for current and future UNICOS releases. CRAY representatives will discuss development plans for UNICOS. Will UNICOS persist beyond SV1, T90, and T3E? What will OS be for SV2? What is overall OS and Programming architecture for SV2? Plans for DMF-like feature for SV2. Transition plans to SV2. Support for LSF and PBS. PC Cluster status. What is CRAY's Professional Services? etc. CRAY representatives will answer these questions and more in a brief presentation of their development plans. Following this 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 CRAY. This meeting will also be an opportunity to express to CRAY what kinds of future conference presentations we would like to have from them.

Introductions of Focus Group Chairs

IRIX, Cheryl Wampler (LANL)

Security, Virginia Bedford and Liam Forbes (ARSC)

LINUX, John Mulholland (CSE)

SGI Liaison, Lynne Johnson
CRAY's future software plans

Open discussion of issues and concerns; wish list discussion
Voting on the top issues and concerns to present to CRAY

7B SIG Open Meeting

The Programming Environments SIG invites you to attend its Open Meeting on Friday afternoon. After a brief introduction of the SIG's business and Focus areas, we discuss 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 Cray Inc. Your feedback on the SIG's business and recommendations for future activities will be gratefully received.

7C SIG Open Meetings

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.

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 Session

Cray Representatives of Cray Hardware Engineering, Software Development and Service organizations will discuss issues and questions about all aspects of Cray Service, Hardware Products and Software Products. The questions will come from

• Questions submitted to the Computer Services SIG chair by email (Leslie Southern, leslie@osc.edu) any time before the CUG meeting.
• Questions placed in the OSC site folder during the CUG meeting. NOTE these need to be submitted by the end of day Thursday 23rd May.
• Questions will also be taken from the floor.

This tends to be a lively session that addresses a wide variety of issues.