A smarter future needs smarter timing
Nasim Samimi Dehkordi defended her PhD thesis at the Department of Electrical Engineering on June 4th.
Autonomous vehicles, smart factories, connected healthcare devices, and advanced robotics all have one thing in common: they must make the right decision at exactly the right moment. A delayed response can be just as dangerous as a wrong one. In her PhD thesis Nasim Samimi Dehkordi introduced new techniques that help these intelligent systems remain predictable, reliable, and fair鈥攅ven when workloads suddenly spike or computing resources are constantly changing. The research demonstrates that real-time guarantees can successfully be combined with modern edge-cloud and Kubernetes technologies, bringing together the flexibility of cloud computing and the strict timing requirements of safety-critical applications.
Cyber-Physical Systems (CPS) combine software with physical processes. Examples include self-driving cars, industrial robots, aircraft control systems, and medical devices. In these environments, software doesn't just need to produce the correct result鈥攊t must do so within a strict time limit. These systems are known as Real-Time Cyber-Physical Systems (RT-CPS). Missing a deadline can reduce performance, compromise safety, or even cause system failures. As CPS become more autonomous and complex, guaranteeing predictable timing becomes increasingly difficult.
The promise and challenge of the edge-cloud continuum
To handle growing computational demands, many modern CPS distribute their workloads across edge and cloud resources. Edge computing processes data close to where it is generated, reducing communication delays. Cloud computing provides vast computational power and scalability. Together, they form the edge-cloud continuum鈥攁 powerful architecture that promises both responsiveness and flexibility. However, this combination also introduces new challenges. Network delays can fluctuate, edge resources are limited, and applications must compete for shared computing capacity. Maintaining strict real-time guarantees across such a dynamic environment is far from straightforward.
Preventing overload before it happens
One of the thesis's key contributions is a new admission control mechanism that decides, in real time, whether an incoming task can safely be accepted. Rather than allowing every request to enter the system, the approach predicts whether a task can still complete before its deadline. If not, the task is rejected before it can disrupt other running applications. This creates a more predictable system that can maintain timing guarantees even when workloads become highly dynamic.
Fairness during traffic peaks
Real-world systems rarely experience perfectly stable workloads. Instead, they often face sudden bursts of activity. The research introduces an admission strategy based on the so-called m/k-firmness model. This model guarantees that a minimum number of requests continue to be served successfully, even during periods of overload. The result is a system that not only protects critical deadlines but also maintains fairness and service continuity across multiple applications sharing the same infrastructure.
Following applications wherever they go
Modern edge-cloud environments are highly dynamic. Devices move, network conditions change, and available resources fluctuate continuously. To address this, the thesis presents algorithms that can automatically adapt resource allocations, scale services up or down, and even migrate applications between edge and cloud locations when needed. Importantly, these adaptations occur while preserving the timing guarantees required by real-time applications.
Bringing real-time guarantees to Kubernetes
Kubernetes has become the standard platform for deploying cloud-native applications. Yet it was not originally designed for strict real-time workloads. The research of bridges this gap by developing mechanisms that allow Kubernetes to deploy and manage containers using Linux's real-time SCHED_DEADLINE scheduler. This enables containerised applications to receive guaranteed processor reservations and predictable execution behaviour. As a result, developers can use modern cloud-native technologies without sacrificing real-time performance.
Turning theory into practice
Beyond introducing new algorithms, the thesis demonstrates that these ideas can be implemented in realistic deployments. A complete Kubernetes-based framework was developed that combines admission control, workload distribution, and real-time scheduling. The framework shows that per-job deadline guarantees remain achievable even in dynamic, multi-tenant environments where multiple applications compete for resources. This represents an important step toward bringing decades of real-time systems research into today's containerised and cloud-native infrastructures.
Building the foundations of dependable autonomy
As autonomous systems become increasingly integrated into daily life, ensuring predictable behaviour will be just as important as improving intelligence.
This research shows that real-time guarantees and cloud-native flexibility do not have to be competing goals. By combining advanced scheduling, adaptive resource management, and modern orchestration technologies, the work lays the foundation for a new generation of dependable edge-cloud systems.
Whether in autonomous vehicles, industrial automation, healthcare, or future smart infrastructure, the ability to guarantee that critical tasks happen on time may ultimately be what makes intelligent systems truly trustworthy.
Title of PhD thesis: . Supervisors: Dr. Marc Geilen, Prof. Twan Basten, and Dr. Mitra Nasri.