Browse Prior Art Database

Operating System Considerations for Large-Scale MIMD Machines

IP.com Disclosure Number: IPCOM000128239D
Original Publication Date: 1985-Dec-31
Included in the Prior Art Database: 2005-Sep-15
Document File: 19 page(s) / 67K

Publishing Venue

Software Patent Institute

Related People

Jan Edler: AUTHOR [+5]

Abstract

In order to realize the potential of highly parallel shared-memory MIZv1D architectures for solving very large problems, novel challenges must be faced by the system software designer. The operating system must endeavor to utilize all processors fully, without incurring serial bottlenecks during coordination operations. Critical code sections far too short or infrequent to cause performance penalties on today's machines will be of concern on very large machines because the cost of each serial section rises linearly with the - number of processors involved. Further, the control software must provide basic facilities to support a structured and natural style of general-purpose - parallel programming. We present the approaches taken for satisfying these requirements in the NYU Ultracomputet-design. We also describe our current preliminary parallel operating system, derived from Urrlx2, which is currently running on an eight-processor prototype Ultracomputer.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 5% of the total text.

Page 1 of 19

THIS DOCUMENT IS AN APPROXIMATE REPRESENTATION OF THE ORIGINAL.

Operating System Considerations for Large-Scale MIMD Machines

Jan Edler, Allan Gottlieb; Jim Lipkis

Courant Institute of Mathematical Sciences New York University 251 Mercer Street . New York, NY 10012

IJltracomputer Dote #92

December, 1985

ABSTRACT

In order to realize the potential of highly parallel shared-memory MIZv1D architectures for solving very large problems, novel challenges must be faced by the system software designer. The operating system must endeavor to utilize all processors fully, without incurring serial bottlenecks during coordination operations. Critical code sections far too short or infrequent to cause performance penalties on today's machines will be of concern on very large machines because the cost of each serial section rises linearly with the - number of processors involved. Further, the control software must provide basic facilities to support a structured and natural style of general-purpose - parallel programming. We present the approaches taken for satisfying these requirements in the NYU Ultracomputet-design. We also describe our current preliminary parallel operating system, derived from Urrlx2, which is currently running on an eight-processor prototype Ultracomputer.

I. Introduction

Continuing advances in microelectronics will produce, during the current decade, megabit memory chips as well as high speed processor chip sets. With 10-20 MIPS (including floating point) and a megabyte soon to be -available on a dozen chips, one is led to ~ consider an extraordinarily powerful machine composed of thousands of processors and gigabytes of memory. Such a configuration would yield several orders of magnitude more performance than do current

'This work was supported in part by the Applied Mathematical Sciences subprogram of the Office of Energy Research, U. S. Department of Energy, under contract number DE- AC0;76ER03077, and in part by the National Science Foundation, under grant number DCR- 8413359. ZUNIX is a trademark of AT&T Bell Laboratories. supercomputers from roughly the same component count. Moreover, the number of distinct components would be quite small and thus the design appears attractive. However, it remains to be demonstrated that this potentially high-performance assemblage can be effectively utilized. There are two dimensions to this challenge. First, several thousand processors must be coordinated in such a way that their aggregate power is applied to useful computation. Serial procedures in which one processor works while the others wait become bottlenecks that drastically reduce the power obtained. Indeed, for any highly parallel architecture, the relative cost of a serial bottleneck rises linearly with the number . of processors involved. Second, the machine must be programmable by

New York University Page 1 Dec 31, 1985

Page 2 of 19

Operating System Considerations for Large-Scale MIMD Machines

humans. Effective use of...