Commodity clusters: Performance comparison between PC`s and workstations

Conference · Fri Mar 01 04:00:00 EST 1996

OSTI ID:251351

Carter, R; Laroco, J; Armstrong, R

Workstation clusters were originally developed as a way to leverage the better cost basis of UNIX workstations to perform computations previously handled only by relatively more expensive supercomputers. Commodity workstation clusters take this evolutionary process one step further by replacing equivalent proprietary workstation functionality with less expensive PC technology. As PC technology encroaches on proprietary UNIX workstation vendor markets, these vendors will see a declining share of the overall market. As technology advances continue, the ability to upgrade a workstations performance plays a large role in cost analysis. For example, a major upgrade to a typical UNIX workstation means replacing the whole machine. As major revisions to the UNIX vendor`s product line come out, brand new systems are introduced. IBM compatibles, however, are modular by design, and nothing need to be replaced except the components that are truly improved. The DAISy cluster, for example, is about to undergo a major upgrade from 90MHz Pentiums to 200MHz Pentium Pros. All of the memory -- the system`s largest expense -- and disks, power supply, etc., can be reused. As a result, commodity workstation clusters ought to gain an increasingly large share of the distributed computing market.

• Research Organization: Sandia National Labs., Livermore, CA (United States)
• Sponsoring Organization: USDOE, Washington, DC (United States)
• DOE Contract Number: AC04-94AL85000
• OSTI ID: 251351
• Report Number(s): SAND--96-8536C; CONF-960835--1; ON: DE96006816
• Country of Publication: United States
• Language: English

Related Subjects

99 GENERAL AND MISCELLANEOUS
BENCHMARKS
COMPARATIVE EVALUATIONS
COMPUTER NETWORKS
DATA TRANSMISSION
DISTRIBUTED DATA PROCESSING
EXECUTIVE CODES
PERFORMANCE
PERSONAL COMPUTERS
SUPERCOMPUTERS