Birds-of-a-Feather sessions, or BOFs, are informal get-togethers for conference attendees to discuss topics of mutual interest. Meeting rooms are provided both before and after the formal hours of the technical program. A BOF notice board with final dates, times, room numbers and any additional sessions will be posted daily in the convention center. Descriptions for the SC2000 BOFs are linked to the titles below.
RAJEEV THAKUR, ARGONNE NATIONAL LABORATORY
JOHN MAY, LAWRENCE LIVERMORE NATIONAL LABORATORY
|TUESDAY NOVEMBER 7|
High Speed Interconnects: Status and Peek into the Future
Organizer: Markus Fischer, University of Mannheim
Time: Tuesday, 5:30 - 7:00 PM
Cluster Computing is one of the most important research platforms for parallel computing in recent years. While standard processors offer high computation performance at low cost, the performance gap to IO is steadily increasing. While Gigabit/s media exist to transfer data, the current bottleneck is the interface between CPU and NIC. This is about to change with InfiniBand(TM) The presentation of the InfiniBandA(TM) specification will be the starting point for a fruitful discussion of the future of IO. Other High Speed Interconnects will be presented, giving an overview of their peek into the future. This BOF will be the continuation of the big success of the last BOF at SC99.
Topics of interest:
Using Clusters for Visualization
As simulation and modeling moves to cluster-based technology, visualization is also being performed on clusters. Come and share the progress everyone has made since the BoF last year. Topics of interest include parallel rendering, panoramic virtual reality, tiled screen displays, cross-platform peripherals, etc.
OpenMP and its Future Developments
OpenMP is the standard application-programming interface (API) for writing parallel programs for shared memory computers. It "was developed" and "is maintained" by the OpenMP Architecture Review Board - or the ARB for short. The goal of the ARB is to promote OpenMP and make sure it evolves to meet the changing needs of the parallel programming community. This BOF provides a major forum for discussion between members of the ARB and the parallel programming community. We have three goals for the session: (1) to present our latest language specification - OpenMP 2.0 for Fortran, (2) discuss our plans for OpenMP 2.0 for C/C++, and (3) hold an open discussion on the future of OpenMP. Members of the ARB will be at the BOF as will vendors who supply OpenMP products. If you have opinions about OpenMP and how it should evolve, come to this BOF and share them.
The ASCI Simulation and Computer Science Technology Prospectus
ASCI's technology prospectus is a set of comprehensive strategic roadmaps (previously known as "ASCI curves") driving ASCI research and development in computational and computer science. These representations depict planned progress and critical barriers that must be overcome for ASCI to meet Stockpile Stewardship objectives. Examples of technology areas include visualization, networking, storage, interconnection networks for ultra-scale computers, and simulation development environments. These roadmaps were developed through a multi-laboratory process where ASCI's experience with applications and platforms efforts led to a clearer understanding of new technology requirements. Furthermore, the introduction of several new ASCI programs, focusing on distance computing, visualization, and data management, required a comprehensive integrated approach to manage interrelated programs. This technology prospectus is an essential part of that integration strategy. This session will provide a forum for presenting and discussing some of these strategic roadmaps that up until now have mostly been developed within the internal ASCI community.
Windows 2000 High Peformance Computing Users BOF
Windows 2000 clusters are rapidly gaining acceptance as a practical way to do high performance computation science on industry standard hardware and software. Already this year NCSA and the Cornell Theory Center have demonstrated that Windows 2000 is scalable and reliable to perform up to the standards of traditionally big iron solutions by making the Top 500 Super Computer List (see http://www.top500.org). A lot of work has gone into developing the necessary tools and programming environments to make the transition from Unix-based systems over to the Windows 2000 environment. In this BOF we will focus on administration experiences and user experiences on Windows 2000 clusters. Topics will include compilers, tools, systems management tools, user experiences, and performance results. Experts from NCSA, CTC, UCSD and others will be on hand to share their experiences. A number of cluster hardware, network and software vendors are also expected to be in attendance.
The ACTS Toolkit Development
The ACTS (Advanced Computational Testing and Simulation) Toolkit is a set of tools mainly developed at DOE labs that make it easier for programmers to write high performance scientific applications for parallel computers. The purpose of this BOF is to gather the tool developers and discuss past and current activities under ACTS as well as future services to the scientific computing community. Feedback from other SC2000 attendees is also welcome.
The ACTS tools have been successfully used by many scientists to address their computational needs. The ACTS toolkit actively promotes High Performance Computing and has provided a framework for educating users of the HPC technology, and we are eagerly working to reach out to other areas in computational sciences. The issues to be discussed during the proposed BOF are tool interoperability, tool evaluation, user education, the on-line information center, success stories, feedback from users and future research under the ACTS umbrella.
Using & Administering GPFS
GPFS is IBM's high performance parallel file system. It is notable for it's scalability and performance: file systems of over 10 TBytes with bandwidths of several GBytes/sec to a single file are deployed and in daily production use. This BOF will provide an informal forum for both scientific app writers and system administrators to discuss issues and findings relevant to using GPFS. Information on the newest release (version 1.3 which was released in July 2000) will be presented. Come with your experiences and/or questions!
Web100: Automatically Tuning TCP for High Performance Networks
Unless the TCP transmit buffer is hand-tuned, TCP sessions across long distance, high-performance networks routinely yield extremely poor results for bulk data transfer. The Web100 project seeks to automatically tune the TCP transmit buffer using congestion feedback information available in real time from the TCP stack .
Tuning algorithms will be implemented at the user level, using a newly defined TCP-MIB that continuously monitors the progress of individual TCP sessions via ab out 3 dozen metrics, and is made available to user processes via an API. Such information may be read by any process, thereby providing a basis for developing TCP diagnostic and monitoring tools. However, processes with root privilege will be able to alter TCP tuning parameters such as each TCP session's TCP transmit buffer.
Minority Serving Institutions (MSI)
The purpose of this project is to increase the participation of minorities in the SC conferences by providing support to help faculty and/or IT professionals from Minority Serving Institutions (MSIs) - Hispanic Serving Institutions (HSIs), Tribal Colleges, and Historically Black Colleges and Universities (HBCUs) to attend the conference. This project takes place in the context of EOT-PACI (www.eot.org) outreach to MSI's under the NSF sponsored Advanced Networking with Minority Serving Institutions (AN-MSI, www.anmsi.org).
The intended impact of this project goes beyond just increasing the numbers of MSI participants in SC conferences to fostering collaborative relationships between faculty at MSIs and research scientists at major research centers and impacting the amount and quality of computational science education at MSIs.
|WEDNESDAY NOVEMBER 8|
Cluster Computing and the IEEE Task Force on Cluster Computing
Organizer: Tim Mattson, Intel Corp.
Time: Wednesday, 5:30 - 7:00 PM
The TFCC is an international forum that promotes cluster computing in research, industry and education. You can learn more about this group at http://www.ieeetfcc.org. In this BOF, we briefly explore the activities of the TFCC. We then move on to a discussion of the key technical issues in modern cluster computing. The format will be informal and consist of panels of experts "on stage" and "in the audience." Topics include programming environments, algorithms and the high performance networks required to build effective clusters.
The TOP500 is a project tracking supercomputer installations in the world since 1993. The sixteenth TOP500 list will be published in November 2000 in time for SC2000. Various experts will present detailed analyses of the TOP500 and discuss the changes in the HPC marketplace during the last years. The number of systems installed, the installed performance, the locations of the various supercomputers, the architectures of HPC systems, and also the applications for which these HPC systems are used, will be analyzed. Such statistics provide a better understanding of the high performance market and can facilitate the exchange of data and software. This BoF will also give background information and explanation about the TOP500 project itself and is also meant as an open forum for discussion and feedback between the TOP500 authors and the user community.
Portable Batch System (PBS) Birds-of-a-Feather
Veridian invites all current and potential users of the Portable Batch System (PBS) to this Birds-of-a-Feather session. Join the PBS support and development staff for a discussion of OpenPBS(tm) v2.3 and PBS Pro(tm) 5.0 capabilities and features as well as planned product enhancements. Share your ideas and experiences with other PBS users. The Portable Batch System, originally developed for NASA, is the leading workload management system for supercomputers and the de facto standard for linux clusters.
Promoting High-Performance Computing Literacy at Low Cost
Many organizations have potential interest in HPC, but lack the physical and intellectual infrastructure to enter this realm. Collaborators at the Arctic Region Supercomputing Center, North Carolina Supercomputer Center, University of Texas Health Sciences Center at San Antonio, and University of Montana are launching a promotional program, targeted at researchers in various disciplines who may not have immediate access to HPC resources. This program will emphasize training and development at a grassroots level, on low-cost clusters with a primary goal of helping researchers "get started" in HPC. It is important that these activities take place in a "portable" manner so researchers may seamlessly move their activities from clusters to supercomputers, and vice versa. The organizers of this program request your presence at this initial BOF, so that we can better understand what you, the researchers, need from us in order to accomplish these goals.
Global Grid Forum
The Grid Forum is a community-initiated forum of individual researchers and practitioners working on distributed computing, or "grid" technologies. Grid Forum focuses on the promotion and development of Grid technologies and applications via the development and documentation of "best practices," implementation guidelines, and standards with an emphasis on rough consensus and running code.
Wide-area distributed computing, or "grid" technologies, provide the foundation to a number of large-scale efforts utilizing the global Internet to build distributed computing and communications infrastructures. As common Grid services and interoperable components emerge, the difficulty in undertaking these large-scale efforts will be greatly reduced and, as importantly, the resulting systems will better support interoperation.
Initially a primarily North American effort, Grid Forum is joining with similar efforts in Europe (www.egrid.org) and Asia-Pacific to form the Global Grid Forum. This BOF will provide an update on Grid Forum activities and objectives.
SCICOMP, The IBM SP Scientific Computing User Group
The IBM SP Scientific Computing User Group, SCICOMP, is an international organization of scientific/technical users of IBM SP systems. The purpose of SCICOMP is to share information on software tools and techniques for developing scientific applications that achieve maximum performance and scalability on SP systems. The SC2000 SCICOMP BOF will review the organization's recently drafted by-laws as well as terchnical concerns raised during the August SCICOMP meeting at SDSC. In addition, SP users and IBM staff will lead technical discussions on topics such as future SP directions and advanced development efforts by IBM's Advanced Computing Technology Center.
HPC Asia 2001
The HPC Asia 2001 Conference and Exhibition will be held 24-28 September 2001 at the Gold Coast in Australia. It is being hosted by the Australian Partnership for Advanced Computing (APAC) and the Queensland Parallel Supercomputing Foundation (QPSF). HPC ASIA is an international conference series held approximately every 18 months in an Asia-Pacific country. It provides a forum for HPC researchers, developers and users throughout the world to exchange ideas, experiences, and research results related to high-performance computing and communications. The first four conferences were held in Taipei in 1995, Seoul in 1997, Singapore in 1998, and Beijing in May 2000. The BOF will discuss conference arrangements and provide information on how you can participate in the conference. Members of the HPC Asia Steering Committee are being invited to attend.
Building Leadership in Computing Fields
Join in an effort shared by women from research centers, labs, and academic departments across North America who are actively addressing the inequities in representation of women in senior management positions in science and technology. To achieve their goal of increased representation of women in all ranks of scientific and technical management, this collaborative effort of women will first obtain information on demographics and succcession plans for US federal agencies involved in high performance computing. In addition to selecting social scientists to conduct a survey of participating organizations' present practices, workshop organizers are actively soliciting community input. This BOF will serve as a forum to collect input and ideas that will improve our study. We will discuss the need for organizations to broaden the definition of leadership, develop best practices to cultivate women leaders, include diversity in succession plans, and acknowledge the importance of pipeline issues. A similar forum was held in September at the Grace Hopper Celebration of Women in Computing.
Visual Serving BOF
Visual Serving solutions provide applications transparent access to high performance visualization and computing resources. They enable interactive visualization of remotely generated supercomputing results, thus increasing individual productivity and resource utilization. This BOF will provide an informal forum for users of SGI's OpenGL Vizserver to meet with the developers and other users to discuss issues and share ideas. Future developments in the OpenGL Vizserver will be discussed and demonstrated.
|THURSDAY NOVEMBER 9|
Discussing a Standardized Hardware Profiling Interface
Organizer: Philip Mucci, UTK/IBM ACTC
Time: Thursday, 5:30 - 7:00 PM
This BOF is to discuss the needs of the community at large for a formalized standard to interface the performance monitoring hardware found on all microprocessors. This hardware can take many forms, the most simple of which is aggregate selectable event counters. Other facilities include instruction tracing and SVR4 sample based profiling. The needs for a formal standard spans the entire community from realtime embedded systems to those systems at the ASCI level and beyond. Current proposed to this solution include PCL from Juelich and PAPI from UTK. The overwhelming response to both projects is further evidence that a more focused an consolidated effort could benefit the computing community as a whole.
The PET Program Training and Outreach to the MSRC and DC Users
PET is the DoD High Performance Computing and Modernization Program (HPCMP) component that is charged with outreach and training of new technologies, to the four Major Shared Resource Centers (MSRCs) and to the multiple Distributed Centers (DCs).
In this forum, we will discuss our joint training efforts, the needs of the DCs, training technologies, and near term training plans. We invite any of those persons involved in training planning and execution from the MSRCs and the DCs to attend.
Cluster System Administration and Management
Clusters of low-cost commodity processors can provide large amounts of cpu cycles at relatively low hardware costs. System administration and management of clusters can be a challenge, however, adding large "human" costs to the system. Please join us for a discussion of cluster system administration and management! We'd like to hear what problems do you run into managing clusters, and what solutions you have found effective.
There are a number (if not many) Open Source projects in this area, but none as yet appear to have captured the community's enthusiasm. Why not? Without enough Open Source momentum, vendors may address cluster system administration and management with proprietary solutions. Let's talk about what Open Source solutions might serve as the basis for a complete cluster system administration and management solution, and what we can do to promote Open Source solutions.
Symbiotic Measurement and Simulation Application Systems
This presentation will discuss a new paradigm for applications, where the application simulations that can dynamically accept and respond to "on-line" field-data and measurements, and/or can control such measurements. This synergistic and symbiotic feedback control-loop between simulations and measurements creates a new generation of dynamic/adaptive applications which are more powerful than the traditional ones where the simulations are conducted with static data inputs. This novel technical direction can open new domains in the capabilities of simulations with high potential pay-off, and create applications with new and enhanced capabilities. It has the potential to transform the way science and engineering are done, and induce a major beneficial impact in the way many functions in our society are conducted, such as manufacturing, commerce, transportation, hazard prediction/management, and medicine, to name a few. The presentation will examine the technical challenges and research areas that need to be fostered to enable such capabilities. Such as: the requirements in the applications' level for enabling this kind of dynamic feedback and control loop; the requirements in the applications' algorithms for the algorithms to be amenable to perturbations by the dynamic data inputs; the challenges and technology needed at the computer systems areas to support such environments. The new set of applications will create a rich set of new challenges and new class of problems for the applications and systems' researchers to address. Such challenges clearly present the need for a synergistic multidisciplinary research between applications and systems' and algorithms' areas. The technical and programmatic issues were discussed in a recent NSF workshop, with participants from academe, industry and several government agencies. The presentation will provide a summary of the workshop report.