HARNESS is an experimental metacomputing system that supports dynamic software reconfiguration, both of the resources that comprise the virtual machine and the services provided therein. Plugins, or distributed modules with certain properties, are the mainstay of this computing model. The plugin architecture greatly enhances reconfigurability, but it also aids failure resilience by providing a framework for the re-instantiation of some types of modules on surviving resources in the event of failures. This paper describes a recent scheme that consolidates this model of failure recovery and enables HARNESS to function as a heterogeneous distributed system that is resilient to a number of types of failures.
Mauro Migliardi, Vaidy S. Sunderam