dtj v07-03 1995

This issue of the Digital Technical Journal (Volume 7, Number 3, 1995) is dedicated to advancements in high-performance computing, particularly for scientific and technical applications, and primarily features two main sections:

1. High Performance Fortran in Parallel Environments: This section details Digital's DEC Fortran 90 compiler, which implements High Performance Fortran (HPF) version 1.1. The compiler generates efficient code for distributed-memory systems, specifically workstation clusters powered by Digital's Alpha microprocessors, focusing on optimizing interprocessor communication. The Parallel Software Environment (PSE) is also introduced as the run-time support for these distributed systems, providing tools for developing, debugging, and profiling parallel HPF applications, and supporting various message-passing protocols (TCP/IP, UDP/IP, shared memory).

2. Sequoia 2000 Research: This larger section covers a multi-year collaborative project (University of California, Digital, government, and industry partners) aimed at building a robust, distributed computing environment for earth scientists studying global change with terabyte-scale data. Key research areas and efforts discussed include:

   • Overall Architecture: An overview of the four-level architecture (footprint, file system, DBMS, application layer) designed to handle high-performance I/O for massive datasets, integrate data into a database management system (initially POSTGRES, later Illustra), and provide advanced visualization and high-speed networking.
   • Electronic Repository: Focuses on the development of text indexing and retrieval software integrated into the POSTGRES database, including methods for probabilistic indexing (Lassen), automatic georeferencing of text documents (GIPSY), and subtopic identification (TextTiling).
   • Visualization Platform (Tecate): Describes "Tecate," a software platform enabling end-users to browse and visualize data from networked sources (databases, World Wide Web). It uses an interpretive, object-oriented language (Abstract Visualization Language, AVL) and an intelligent system to automatically craft interactive visualizations in virtual worlds.
   • High-Performance I/O and Networking Software: Addresses the critical need for efficient movement of very large data objects across networks. Solutions include novel operating system I/O software models like "container shipping" and "peer-to-peer I/O" to avoid costly memory copying and reduce system overhead, along with optimizations for TCP/IP and UDP/IP network protocols by reducing checksum computation.

The journal emphasizes the importance of treating such large-scale projects as "end-to-end solutions" and highlights lessons learned regarding infrastructure, interdisciplinary collaboration, and the practical challenges of deploying research prototypes.