Application transparent fault management in fault tolerant Mach

A general purpose operating system fault management mechanism, the sentry, has been defined and implemented for the Mach 3.0 microkernel running a UNIX 4.3 BSD server. The value of a mechanism in the operating system domain is usually judged by two criteria: the suitability of the mechanism to support a wide range of policies and the performance cost of the mechanism. Similarly, in fault detection and recovery there are a relatively large number of strategies which can be mapped onto mechanisms and policies for fault tolerance. To highlight the properties of the sentry mechanism for fault management, the suitability and performance of the proposed mechanism are being evaluated for sample fault detection policies and for sample fault recovery policies. In the fault detection domain use of the mechanism to support assertion type policy is presented and evaluated through an example. Two recovery policies have been chosen and evaluated: checkpoint/restart and checkpoint/restart/journaling.