A multiple computer system for reliable transaction processing
John V. Levy · ACM SIGSMALL Newsletter · 1978
The Tandem 16 system is a novel solution to the problem of providing a cost-effective and reliable computer system for commercial transaction processing. The system design goal is to continue operation in spite of any single component failure.This paper presents an overview of the hardware structure of the system. It is a multiple computer system, with no shared main memory. The software structure that implements "NonStop" (or fault-tolerant) computing is then described. The key concept is "paired processes". For every software module that is to run "NonStop", there is a backup copy sleeping in an alternate processor module. On detection of a failure, the backup copy is activated, and it takes over processing at the last checkpoint. All system processes are fault-tolerant, and considerable care is taken to avoid duplication of I/O operations during recovery.