The Reliability of Ground-Based Digital Computers
Abstract
This Memorandum discusses the many aspects (both qualitative and quantitative) of obtaining a reliable digital computer and, in particular, investigates that class of ground-based data processing systems where repair is possible. The study begins by reviewing the reliability of computer parts (transistors, capacitors, integrated circuits) and applies the results to a large variety of probabilistic models of system availability. Further, it discusses the availability of ground-based data processing systems-- specifically, the probability that a repairable computer which should be ready will in fact be ready for use at some arbitrary future time. It is concluded that part failure distributions show a form of decreasing failure rates for the entire population which in no way correlates with the predicted behavior of the ideal part. The total part population shows a decreasing failure rate because, and only because, various controllably small subgroups show initially increasing failure rates until every member of the subgroup has failed, at which time the failure rate of the entire population effectively decreases. Next, the authors survey the various machine structures which yield higher reliability and show that, where service is available, redundancy is never a contender as a means to high reliability. With service, the multi-processor (duplex or triplex), second highest.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 01, 1965
- Accession Number
- AD0467808
Entities
People
- Michael Warshaw
- Rodger R. Lowe
Organizations
- RAND Corporation