Design and implementation of a distributed monitor facility

Abstract

There are four major drawbacks to use existing mechanisms, such as which are provided by Mach and Masscomp, etc., for designing monitors. (1) They cannot support fault-tolerance. (2) They are tailor-made for specific kernels. (3) They cannot directly be applied in distributed environments. (4) Users must have good backgrounds in handling light-weight processes. In this paper, we present a distributed monitor facility, called DISMON, that supports the mechanisms needed to design monitors and makes them fault tolerant to hardware failures. At the same time, monitors are constructed at the application level not kernel level such that they are portable and easily designed. In order to produce a monitor, the source monitor program, which makes use of a C-like language to code it in the DISMON, is pre-compiled by a monitor generation program, called MONGEN, to generate it. The monitor procedure calls are made entirely transparent by means of an agent based name server that helps callers to obtain the address of a monitor. The interface between callers and monitor are similar to the Sun's Remote Procedure Call (RPC) mechanism, but the RPC servers have been replaced by monitors.