From: jgreen@Alliant.COM (John C Green Jr) Subject: Alliant Announces CAMPUS: An 800 CPU, 32 GFLOP, 128 GByte MPP System Date: 23 Nov 91 06:55:23 GMT Long Press Release Follows: ALLIANT INTRODUCES FIRST MASSIVELY PARALLEL SUPERCOMPUTER DESIGNED FOR EASY PROGRAMMING AND COMPUTATIONAL EFFICIENCY CAMPUS/800 Brings MPP to Mainstream Users; Based on Alliant's Proven FX/Series Technology Albuquerque, NM, November 18, 1991 -- Alliant Computer Systems today introduced the CAMPUS/800, a massively parallel processing (MPP) supercomputer with up to 800 standard RISC processors. The Alliant CAMPUS/800 is the first system to support both distributed and shared memory, making it easier to program and more efficient for real-world applications. Current MPP systems--such as those from Thinking Machines and Intel SSD--are based exclusively on distributed memory, in which messages must be passed back and forth between large numbers of processors. This increases programming complexity and severely limits efficient use of processor resources. However, distributed memory is required to build MPP systems which `scale up' to hundreds of processors. Shared memory, on the other hand, allows processors to easily and rapidly share data and coordinate tasks. This makes programming easier and yields greater application performance to the end user. The CAMPUS/800 combines the programming simplicity and computational efficiency of shared memory with the scalability of distributed memory, resulting in an MPP system that is significantly more usable than current MPP offerings. In addition, the CAMPUS two-tier memory architecture better matches the natural structure of most large-scale problems, which typically exhibit both fine-grain and coarse-grain parallelism. The CAMPUS architecture uses shared memory for fine-grain parallelism and distributed memory for coarse-grain parallelism. `Our experience has shown that a two-level memory architecture is the most natural and computationally efficient mechanism for utilizing large numbers of processors in a massively parallel system,' said Dr. David Kuck, professor and director of the University of Illinois' Center for Supercomputer Research and Development. `CAMPUS is a natural extension to the collaborative parallel architecture development at CSRD and Alliant over the past 10 years.' Massively parallel systems are used to solve large scale problems which cannot be effectively handled by traditional supercomputers. However, MPP systems have been difficult to use compared to these traditional systems. As a result, they have been little more than fast, exotic machines used primarily for experiments by computer scientists rather than to solve real-world problems. `The CAMPUS/800 supercomputer addresses an urgent need for a massively parallel system that can be used in production-level environments,' said Craig Mundie, Alliant's president and CEO. `Experts agree that MPP is needed to bring about significant leaps in supercomputer performance. But to date, MPP systems have lacked many key usability features demanded by mainstream computer users. CAMPUS is the first system to combine MPP performance with the usability that technical users have come to expect.' CAMPUS provides parallel programming tools for productivity, standard third-party applications for ease-of-use, and a versatile UNIX operating system environment for accessibility. When combined with the benefits of the CAMPUS two-tier architecture, these make the system immediately available to mainstream supercomputer users--engineers and scientists who use supercomputers for real-world applications such as designing new drugs or exploring for oil. The CAMPUS/800 MPP system provides up to 32 peak GFLOPS of computing power and 128 Gbytes of memory. A fully configured CAMPUS/800 system is composed of 32 supercomputer ClusterNodes with each `node' consisting of a `cluster' of 25 Intel 64-bit i860 RISC processors and 4 Gbytes of high-speed shared memory. For rapid data sharing and synchronization, the ClusterNodes communicate over a 2.56 Gbyte/s High-speed Memory Interconnect (HMI). Each ClusterNode is based on the proven FX/2800 supercomputer, enhanced via hardware and software for high-performance MPP operation. CAMPUS can be used as a single large MPP system for extremely large problems or as multiple autonomous supercomputers for individual departments. Advanced operating software dynamically partitions the CAMPUS system based on user requirements and changing workloads. High-Productivity Parallel Tools Ease-of-programming is provided by a sophisticated set of development tools, including Alliant's automatic parallel compilers and libraries--based on nearly 10 years of experience in producing parallel processing software--and the high-level EXPRESS environment from ParaSoft Corporation (Pasadena, CA). Other standard software tools for CAMPUS include TCGMSG and p4 (from Argonne National Laboratory) and PICL (from Oak Ridge National Laboratory). Immediate Availability of Third-Party Applications Typical applications for the CAMPUS system include image and signal processing, seismic processing, computational chemistry, computational fluid dynamics, and structural analysis. Over 100 third-party applications are available to run on the parallel ClusterNode, having previously been developed and delivered on the existing FX/2800 supercomputer. This allows users to be productive immediately and then gradually move applications to larger numbers of processors over time. Alliant is also porting several applications and benchmarks for MPP operation over multiple ClusterNodes. By comparison, very few off-the-shelf applications are available for competitive systems because of the extensive modifications and effort required by experts. Versatile Operating Environment CAMPUS is the first MPP system with a full-function UNIX operating system executing on each node of the MPP system itself. Other MPP vendors rely on a separate workstation or VAX `front-end' to execute the operating system and provide services such as time-sharing access, UNIX file I/O, network connectivity and software development facilities. This limits the number of simultaneous users, since access to the system is limited by the power of the front-end, and increases the overall cost of the system. In contrast, the CAMPUS system supports simultaneous execution of computational tasks with UNIX file I/O, network activity and development on each of the ClusterNodes, thus providing a significantly more versatile and robust operating environment and supporting a greater number of users. `We have already successfully moved our existing large chemistry applications to the FX/2800, resulting in comparable performance and much better job turn-around than we had experienced on the Cray YMP,' said Robert Harrison, assistant chemist at the Argonne National Laboratory. `With CAMPUS, these applications can be extended to tackle chemistry problems that are impossible to do with traditional supercomputers.' Availability and Pricing Production availability of CAMPUS/800 systems is scheduled for Q2 1992. Upgrades from current FX/2800 systems have begun shipping this quarter to Alliant customers as part of the system's test and evaluation program. The customers include: Argonne National Laboratory (Argonne, IL) and the National Office for Aerospace Research and Studies (ONERA) in France. The system is priced at $2.5M for a typical configuration with 100 processors. An 800-processor CAMPUS system is priced at $16M. FX/2800 Family systems start at under $200,000 and can be used as fully-compatible CAMPUS `starter nodes.' Upgrades for current FX/2800 systems are also available starting at under $100,000. Alliant Computer Systems Corporation is the leading manufacturer of standards-based parallel supercomputers. More than 650 systems have been installed by customers around the world to enable scientists and engineers to quickly solve complex problems.