| Popis: |
One of the key challenges in dynamic reconfiguration of time-critical distributed embedded applications is state transfer: For consistency, the old component is usually blocked and its state is first extracted, then potentially transmitted and transformed and finally injected to the new component, before it can take over. However, as the technical process needs to be controlled continuously, there is technically no time for blocking. We present an approach to reconfigure stateful distributed embedded applications during operation either without or at least with bounded quality degradation. Using an extended prototype of our real-time container architecture the state transfer can be performed temporally deterministic and coordinated across nodes. By means of the reconfiguration plan the plant operator can employ strategies to keep up the system functionality for a limited time frame while specific components and messages are blocked. We describe a case study with three different reconfigurations showing the applicability of the proposed approach to various scenarios. We also discuss trade-offs and opportunities for future improvement of our approach. |
