A foundation for adaptive fault tolerance in software

Abstract

Software requirements often change during the operational lifetime of deployed systems. To accommodate requirements not conceived during design time, the system must be able to adapt its functionality and behavior. The paper examines a formal model for reconfigurable software processes that permits adaptive fault tolerance by adding or removing specific fault tolerance techniques during runtime. A distributed software-implemented fault tolerance (SIFT) environment for managing user applications has been implemented using ARMOR processes that conform to the formal model of reconfigurability. Because ARMOR processes are reconfigurable, they can tailor the fault tolerance services that they provide to themselves and to the user applications. We describe two fault tolerance techniques: microcheckpointing and assertion checking, that have been incorporated into ARMOR process via reconfigurations to the original ARMOR design. Experimental evaluations of the SIFT environment on a testbed cluster at the Jet Propulsion Laboratory demonstrate the effectiveness of these two fault tolerance techniques in limiting data error propagation among the ARMOR processes. These experiments validate the concept of using an underlying reconfigurable process architecture as the basis for implementing replaceable error detection and recovery services.