Reliable Scheduling in a TMR Database System
A Triple Modular Redundant (TMR) system achieves high reliability by replicating data and all processing at three independent nodes. When TMR is used for database processing all non-faulty computers must execute the same sequence of transactions, and this is ensured by a collection of processes known as schedulers. In this paper we study the implementation of efficient schedulers through analysis of various enhancements such as null transactions and message batching. The schedulers have been implemented in an experimental TMR system and the evaluation results are presented here.