A comparison of predicted MTBFs to field and test data

Tandem Computers manufactures fault-tolerant computers for the on-line transaction processing (OLTP) market. Although the fault-tolerant architecture prevents a single component failure from causing a system failure, reliability is still important to Tandem to decrease their service costs and increase customer satisfaction. Tandem have been performing standard MTBF calculations using handbooks for many years. As has been well documented in the literature, there are many problems associated with using these handbooks. Over the years Tandem found that the numbers were very conservative compared to field data. While they are pleased that the predictions are conservative, Tandem feel it is important to have realistic reliability predictions for: setting reliability goals; evaluating design reliability; life-cycle cost trade-off studies; determining optimal test times; and service planning. In order to provide realistic MTBF estimates or vendor specifications, Tandem needed to determine the differences among MTBFs derived from their predictions, from manufacturing test data, and from field performance data. During 1991 and 1992 they gathered data on a number of their products. This paper describes the results of their study - a comparison among predicted MTBFs, demonstrated MTBFs, and field MTBFs