Cray Research, Inc.

Supercomputer Systems Limited Partnership

Edward C Priest

Steve S Chen

David A Hanson

John M Wastlick

Dan L Massopust

Douglas C Paffel

George A Spix

Douglas R Beard

Frederick J Simmons

Steven S Chen

Marvin D Bausman, Jr

Ken Shintaku

Steven C Barber

Greg W Pautsch

Michael H Rabska

James M VanDyke

Dennis F Girling

Lisa Heid

Felix R Lesmerises

Christopher J Sperry

A unified parallel processing architecture connects together an extendible number of clusters of multiple numbers of processors to create a high performance parallel processing computer system. Multiple processors are grouped together into four or more physically separable clusters, each cluster having a common cluster shared memory that is symmetrically accessible by all of the processors in that cluster; however, only some of the clusters are adjacently interconnected. Clusters are adjacently interconnected to form a floating shared memory if certain memory access conditions relating to relative memory latency and relative data locality can create an effective shared memory parallel programming environment. A shared memory model can be used with programs that can be executed in the cluster shared memory of a single cluster, or in the floating shared memory that is defined across an extended shared memory space comprised of the cluster shared memories of any set of adjacently interconnected clusters. A distributed memory model can be used with any programs that are to be executed in the cluster shared memories of any non-adjacently interconnected clusters. The adjacent interconnection of multiple clusters of processors to a create a floating shared memory effectively combines all three type of memory models, pure shared memory, extended shared memory and distributed shared memory, into a unified parallel processing architecture.

Extendible clock mechanism

An optical clock distribution method and apparatus is disclosed that minimizes clock skew in the distribution of clock signals to logic assemblies in a computer system. The logic assemblies convert the optical signals into equivalent electrical signals.

Optical clock distribution method and apparatus

Method and apparatus for a unified parallel processing architecture

Utilization circuits, such as logic chip circuits, are prevented from receiving the initial one or more pulses of a train of clock pulses produced after the master system clock is started, while the pulses of that train occurring thereafter are coupled to the utilization circuit. This prevents the skew usually present between the initial pulses of the train relative to the subsequent train pulses from adversely effecting operation of the utilization circuits. This clock swallowing preferably blocks a certain predetermined number of initial clock pulses from reaching the rest of the circuitry, although the system is adaptable to allow preselection of the number of such swallowed pulses.

Clock start up stabilization for computer systems

Clock pulses from a master oscillator are distributed in a multiprocessor computer system so that they arrive at a large number of utilization points located in operating clusters of modules within extremely tight time tolerances of each other. The delays associated with each component, electrical or optical connection, cable or the like are determined by direct measurement or by using known standard characteristics. A time delay budget for each complete clock pulse path from the point of initial divergence from the master clock source to the final chip delivery point is logged and summed. Components capable of introducing predetermined amounts of time delay are incorporated in some or all clock pulse paths. These components are adjusted so as to balance out the differences determined from the clock path budgets. The clock paths are implemented in electrical components either alone or in combination with optical components, or in substantially all optical configurations. One arrangement for controlling optical skew includes an arrangement of optical elements physically displaceable in a coaxial direction relative to one another. Skew adjustment networks employ a unique composition of coarse and fine selectable delay arrays implemented either by electrical components, optical components, or a combination thereof.

Clock distribution apparatus and processes particularly useful in

The present invention includes methods and apparatus for creating a packaging architecture for a highly parallel multiprocessor system. The packaging architecture of the present invention can provide for distribution of power, cooling and interconnections at all levels of components in a highly parallel multiprocessor system, while maximizing the number of circuits per unit time within certain operational constraints of such a multiprocessor system.

Packaging architecture for a highly parallel multiprocessor system

Growing community of inventors

Eau Claire, WI, United States of America

Edward C Priest