Evolutionary algorithm for scheduling real-time applications in system of systems

In recent years, systems engineering and management have evolved from developing distributed
systems to the integration of complex adaptive systems and the advent of Systemsof-
Systems (SoS). SoS emerge from the collaboration of multiple systems with operational
and managerial independency in order to accomplish a higher goal. SoS have been successfully
deployed in different domains such as enterprise systems and smart cities. However,
there is a critical challenge that must be tackled in order to adopt SoS in safety-relevant
embedded applications: reliability and real-time capability are today not addressed in SoS.
An open research challenge is the development of a distributed embedded system architecture
for constantly evolving and dynamic SoS with support for verifiable real-time and
reliability properties. The system architecture needs to support reliable closed loop control
with stringent real-time requirements for applications.
Most of the existing scheduling solutions are developed for monolithic systems or complex
systems with centralized authorities, which may violate the restrictions of SoS and
not be able to satisfy its requirements. In this thesis, we develop an efficient heuristic
approach for scheduling SoS applications with real-time and fault-tolerance requirements.
In order to respect the SoS architectural restrictions, we model the scheduling decisions
at two levels using a Genetic Algorithm (GA) optimizer as a solver, which iteratively interact
to reach a feasible and efficient schedule for the SoS. The computational results
show improvement in the average transmission makespan of SoS applications compared
to the state-of-the-art scheduling solutions up to 31 percent in different scale scenarios.
This work also investigates the capability of our scheduling approach in computing timetriggered
schedules for a sequence of incrementally added SoS applications in a real-time
SoS network. In this regard, a heuristic approach is developed at both scheduling levels
to improve the schedulability of our algorithm by efficiently sparing free time slots on resources
for the upcoming applications. Testing the schedulability and timeliness of the new
incremental scheduler on a set of applications shows improvements in schedulability of up
to 50 percent. Furthermore, we design a fault-tolerant scheduling approach for real-time
SoS applications to tolerate permanent faults. Accordingly, fault-tolerance techniques such
as re-execution and replication are integrated into our two-level GA scheduling algorithm
to enhance the reliability of the system in combination with satisfying deadline constraints.
The reliability is improved on average by 15 percent compared to the non fault-tolerant
scheduler in different scenarios.