Master Student Computer Science or similar (f/m/x) - Evaluation of On-line Reconfiguration Techniques for a Distributed Avionic Middleware

More and more space missions use commercial-off-the-shelf (COTS) components in order to provide the necessary computational performance for modern applications, like image processing or autonomy algorithms. Often those COTS components are not space-grade, meaning that they are vulnerable to radiation. Incoming radiation can lead to so called soft errors in those components. These vary in their severity, ranging from no effect up to a complete component failure.

A possible way to overcome the issue of providing safe operations for a spacecraft in the presence of a node-stop failure is the establishment of a distributed architecture in which several interconnected nodes provide the desired functionality for the spacecraft. In case of a failure of such a node, other nodes can take over the responsibilities and tasks of the failed one until it recovered.

A fault-tolerant middleware for a distributed, heterogeneous on-board computer is developed in the context of the Scalable On-Board Computing for Avionics (ScOSA) project. The middleware includes the functionality of distributing the tasks of the spacecraft to the nodes. The middleware does that in a static, pre-defined manner, meaning that all possible scenarios of node failures and task distributions will be calculated at design time and stored in the middleware, such that it can recall them as soon as the corresponding node failure scenario appears during run-time. Handling the reconfiguration in this static way brings the advantage of a fast reaction in case of a failure, while also being easily traceable for human operators on ground. The downside of this approach is the consumption of additional memory to store all possible scenarios even when they will never be used.

The opposite of such a static off-line mechanism is the approach of reconfiguring on-line, which means that the reaction to a node failure will not be pre-calculated but decided in the moment of the failure by the middleware itself. This involves deducting the best possible configuration, given the current state of the system in particular the nodes.

The goal of this thesis is to research, develop and evaluate on-line reconfiguration mechanisms for the ScOSA middleware focusing on the task distribution and the path routing between the nodes. Another important aspect of the researched mechanisms is the traceability for human operators, meaning that decisions made by these algorithms can be easily understood on ground. Additionally, the mechanism should also work in a decentralized way if possible.

The thesis includes the following tasks:

• research of possible on-line reconfiguration mechanisms focusing on traceability
• Implementation of researched techniques into the ScOSA middleware
• Evaluation of the implemented techniques
• comparison to the state-of-the-art off-line mechanism