A system model for dynamically reconfigurable software

The ability to reconfigure software is useful for a variety of reasons, including adapting applications to changing environments, performing on-line software upgrades, and extending base application functionality with additional nonfunctional services. Reconfiguring distributed applications, however, can be difficult in practice because of the dependencies that exist among the processes in the system. This paper formally describes a model for capturing the structure and run-time behavior of a distributed system. The structure is defined by a set of elements containing the state variables in the system. The run-time behavior is defined by threads that execute atomic actions called operations. Operations invoke code blocks to bring about state changes in the system, and these state changes are confined to a single element and thread. By creating input/output signatures based upon the variable access patterns of the code blocks, dataflow dependencies among operations can be derived for a given configuration of the system. Proposed reconfigurations can be evaluated through off-line tests using the formal model to determine whether the new mapping of operations-to-code blocks disrupts existing dataflow dependencies in the system. System administrators—or software components that control adaptivity in autonomic systems—can use the results of these tests to gauge the impact of a proposed reconfiguration on the existing system. The system model presented in this paper underpins the design of reconfigurable ARMOR processes that provide flexible error detection and recovery services to user applications.