scray.jpg (5654 Byte)
Seymour Cray

CUG - A Forum for HPC Users

 

Abstracts of NAS Papers Presented at CUG



Scalability and Sustained Performance per Dollar of CRAY C90, CRAY T90, CRAY J932, and CRAY T3D

Subhash Saini

The performance and architectures of Cray parallel vector processors, highly parallel processors, and symmetric multiprocessors have been evaluated and compared with other supercomputers. In addition, the scalability of numbers of processors and sustained performance per dollar is presented for the following systems: CRAY C90, CRAY T90, and Fujitsu VPP500; CRAY T3D, IBM SP2-Wide Nodes, and IBM SP2-Thin Nodes 2; and Convex Exemplar SPP1000, CRAY J932, DEC Alpha Server 8400 5/300, and SGI Power Challenge XL (90 MHz).

To request a copy of this paper, email saini@nas.nasa.gov.


A Methodology for Calibrating the Cray Research Hardware Performance Monitor: Initial Observations

Eugene Miya (now of Ames' Distributed Systems Division)

The CRAY X-MP/Y-MP Hardware Performance Monitor (HPM) can provide low-overhead instruction counting but is sensitive to hardware and software configuration. Up to now, its software has lacked calibration, but the author concludes that calibration is needed for languages, compilers, and measurement hardware on all vendor measuring tools.

To request a copy of this paper, email eugene@.ames.arc.nasa.gov.


Integrating D3 Technology into a DMF Configuration

Alan K. Powers

The NAS Facility (with its CRAY C916/1024, 260+ gigabytes of Superhome storage, and 2 terabytes of migrated storage) has served as a beta test site for the Storage Technology Corp. (STK) Redwood SD-3 device and Cray's DMF [Data Migration Facility] 2.3. This paper includes an overview of the problems and various metrics for the SD-3 tape drive. In addition, by using the DMF 2.3 advanced tape media-specific process with existing STK 4490 tape drives-and integrating the Redwood drives-the number of 3490E tapes can potentially be reduced by 90 percent.

To request a copy of this paper, email powers@nas.nasa.gov.


Monitoring Object Library Usage and Changes

R.K. Owen

This paper presents an overview of several tools developed for monitoring object library usage and their performance results at the NAS Facility/Aeronautics Consolidated Supercomputing Center. Some tools allow non-invasive monitoring; others implement runtime logging-identifying who, when, and what is being used-even for object-only libraries. Benefits of these tools include being able to measure real library usage, discontinue unused libraries, and apply actual numerical methods to training and optimization efforts.

To request a copy of this paper, email rkowen@nas.nasa.gov.


UNICOS Kernel Internals Application Development

Nicholas Cardo

In order to develop applications that utilize kernel information, an understanding of the kernel internals is necessary. This paper discusses the intricacies of developing utilities that access and use kernel information, such as algorithms and code. Code segments demonstrating how to locate and read kernel structures are provided. In addition, the types of applications that can utilize kernel information are also covered.

To request a copy of this paper, email cardo@nas.nasa.gov.



[Search] [Home] Updated: February 11, 1997
WebWork: wwwmaster
NASA Responsible Official: Lisa Reid

© Cray User Group Inc. All rights reserved.  Page last modified: 16 Jul 03