System calls are a primary way in which applications to communicate with the kernel. This is to allow them to perform sensitive tasks, however, an application will typically not require all of the system calls available to function properly. Despite this, the Linux kernel allows a program to perform any system call it wishes. This is bad for security, as it allows an attacker full access to the kernel after gaining code execution in a vulnerable program. By extracting a minimal set of system calls for a given program, we can sandbox it and only allow those system calls to be executed, greatly reducing the attack surface. In this paper, we analyze existing solutions that address system call set extraction. In particular, we will focus on applying these to configurable binaries. That is, binaries which can be compiled with a variety of different settings. For this paper, we have chosen to analyze cat as a minimal example, and busybox as the configurable application. We compile busybox in the following configurations, among variations: the default configuration, a configuration containing a minimal set of features and a configuration containing a maximal set of features. We analyze the performance of the tools Binalyzer, Sysfilter and Confine on these binaries. We see that Confine has significantly worse performance than both Binalyzer and Sysfilter. We also see that Sysfilter has better performance than Binalyzer when the complexity of the busybox binary is increased. We conclude that Sysfilter outperforms Binalyzer on binaries without debug symbols, while the opposite is true when performing analysis on binaries with debug symbols.

System call sandboxing is the idea to restrict the set of system calls an application is able to invoke.
This reduces the attack surface available to an attacker exploiting the binary, and adheres to the principle of least privilege, giving entities the minimum required permissions needed to perform their function.

The key goal is to automatically identify which system calls to block, since it is a complex, manual task requiring great insight into the program and its dependencies.

This paper investigates and compares various static analysis based solutions in this field, such as sysfilter [3], Confine [5] and Chestnut [2], by measuring their accuracy and analysis time. Furthermore a simple dynamic analysis based solution is created for the sake of comparison with the previously mentioned tools. The tools are evaluated on a small set of commonly used Linux applications, such as ls, sqlite and Redis, and the results are reported.

In addition to the aforementioned tools, temporal specialization [6], a solution which considers multiple execution phases is also investigated and compared with a purely dynamic analysis solution having support for multiple phases of execution.

The research shows that although dynamic analysis underapproximates the set of required system calls it can adapt to a custom usage profile.
Additionally, although static analysis is slower and more complex, the research explores areas of improvement such as precomputing or multiple threads.

Continuous Integration (CI) is a widely used quality-assurance measure within software development. It empowers developers to spot bugs and integration issues early in the development cycle and helps to maintain a coherent codebase, both in terms of quality and styling, even in open-source environments. But CI might have a hidden cost. Projects need to be built and tested continuously throughout the development cycle. It is not uncommon for projects to have multiple commits per day, reaching thousands of commits per year, with each commit having one or multiple build and test cycles. In this thesis, over 200 open-source Java projects were measured with the aim of making developers more aware of how much energy these builds can take and the measures that can be taken to reduce energy consumption where possible.

The averse reaction of lithium batteries inspired researchers to look for alternat-
ives for energy storage in electronics. Battery-free or intermittent devices intro-
duce a new paradigm with their own specific problems about power, state track-
ing, timing and communication. We implemented the battery-free paradigm in
the Thread protocol, utilizing the specific characteristics of End Devices in the
mesh network to reduce power consumption. The intermittent implementation
is fulfilled without changing the behaviour of the protocol, making the device
fully compatible without other devices being aware of the intermittent opera-
tion. During tests four implementations are tested, completely naive, semi naive
with simple power improvements, Sleepy End Device operating intermittently
and a Synchronised Sleepy End Device operating intermittently. The results of
the power measurements are used to determine the feasability of four use cases.
The intermittently operating Sleepy End Device is out performing the Naive
and Semi Naive implementation.

Over development of the past three decades, control systems in automobiles have undergone a significant transformation, shifting from mechanical devices to a multitude of interconnected computers that oversee sensors, drivers, and passengers. In today's modern vehicles, 50-80 independent computers, commonly referred to as Electronic Control Units (ECUs), form complex in-vehicle networks to facilitate communication among themselves. A widely adopted communication protocol in modern in-vehicle networks is the Controller Area Network (CAN). Security wasn't considered when the CAN protocol was designed in the mid-1980s, leaving it susceptible to cyber-attacks. In 2015, Charlie Miller and Chris Valasek successfully demonstrated a remote attack against a Jeep Cherokee, compromising critical components such as the steering wheel and braking system. This event prompted significant concern about the security of CAN networks in both academic and industrial circles.


In reaction to these challenges, researchers have formulated a variety of attacks targeting the CAN bus, encompassing Denial-of-Service (DoS) attacks, spoofing attacks, and more. Furthermore, significant efforts have been directed towards enhancing the security of CAN networks. These endeavors encompass the design of security architectures and intrusion detection systems. NXP developed the TJA115x secure CAN transceiver family, which offers protection against various types of attacks for Classical CAN and CAN FD communication without resorting to cryptography. Nonetheless, all existing researches are directed towards the Classical CAN and CAN FD protocols, leaving a notable gap in the exploration of security aspects concerning the latest CAN XL network.

This thesis focuses on enhancing the design of the TJA115x secure transceiver to support CAN XL protocol. The research comprises two main aspects: flooding detection and flooding prevention. To achieve flooding detection, a leaky buckets tree is introduced, enabling differentiation between normal frames and flooding frames, along with dynamic traffic control for each frame type. Additionally, the leaky bucket parameters are derived in formulaic form. For flooding prevention, two distinct strategies are devised to prevent flooding frames from dominating the bus. These strategies are compared in terms of throughput, latency, and availability. Finally, the proposed secure CAN XL transceiver is validated through software simulations and hardware experiments. The results demonstrate the exceptional performance of the CAN XL secure transceiver in blocking flooding attacks and safeguarding communications. These accomplishments are achieved with minimal memory overhead, and notably, without introducing any additional latency or bandwidth demands.

The wake effect which is turbulence behind a wind turbine created when it extracts energy negatively impacts the power output of the downstream turbines. Active Wake Control can mitigate this effect, by rotating some turbines away from the wind. Previous research applied single agent reinforcement learning to apply Active Wake Control, show- ing good results for small-scale layouts, that don’t scale for larger, practical wind farms.
To that extent, this study focuses on the application of mean-field multi-agent reinforcement learning to Active Wake Control, under constant wind conditions. This algorithm limits the computations to a limited set of neighbouring turbines, reducing their complexities. To build the answer to this question I will also study:
1. how to model the rewards to solve the lazy- agent problem, leveraging the nature of the Active Wake Control
2. how the view of the agent changes the results
3. how does it compare to a single-agent reinforcement learning algorithm, TD3
The experiments were done using the Floris Wake Simulator, with each turbine sharing the same agent, placed in tunnel layouts at real-life distances (6-7 rotor diameters), under constant wind conditions.
Results show that with the proper configuration of rewards and view space within wind tunnels, the mean-field algorithm finds near optimal configurations for Active Wake Control, within a small number of episodes. This shows a promising start for the application of mean-field multi-agent algorithms for the Active Wake Control problem, and provides insight into how to model the rewards, which might be applicable for the whole class of algorithms.

The close proximity of wind turbines to one another in a wind farm can lead to inefficiency in terms of power production due to wake effects. One technique to mitigate the losses is to veer from their individual optimal direction. As such, the wakes can be steered away from downstream turbines in order to increase the overall power output. Multi-Agent Reinforcement Learning (MARL) models the interactions between wind turbines and determines an optimal control strategy through agents that learn the collective consequences of their actions. To
analyse the benefit of multi-agent cooperation and centralised critic evaluation, I investigated the effect of Counterfactual Multi-Agent Policy Gradients (COMA) on Active Wake Control. Ultimately, experiments on wind farms of three and sixteen turbines indicate that the algorithm performs moderately, yet worse than single-agent Reinforcement Learning. In addition, high computation costs hinder its application on real-life environments.

Wind energy, generated by windfarms, is playing an increasingly critical role in meeting current and future energy demands. windfarms, however, face a challenge due to the inherent flaw of wake-induced power losses when turbines are located in close proximity. Wakes, characterized by regions of turbulence and lower wind speed, are created as air passes through the rotors, reducing the efficiency of downstream turbines. Wake losses can be reduced by yawing upstream turbines to steer the wake away from downstream turbines. While there are power losses associated with turning turbines off-wind, the gains in the subsequent turbines can outweigh these losses. Yawing turbines to increase overall power output is known as Active Wake Control, and the literature shows that single-agent reinforcement learning algorithms can be used to learn such control policies. However, these approaches are limited to a small number of turbines and do not scale well to larger windfarms. Multi agent reinforcement learning algorithms do scale to larger windfarms and this paper investigates the application of the DGN algorithm for windfarm active wake control. DGN is a fully cooperative algorithm that utilizes a graph representation of agents to encourage collaboration among neighboring turbines. The use of DGN is particularly interesting because windfarms naturally have a topological structure, and depending on the modeling choices, graphs can capture a significant amount of information. This paper demonstrates that DGN can learn useful policies for wake control. Although it does not outperform the DQN single-agent algorithm on small windfarms, its advantage becomes apparent in larger windfarms, where its performance remains consistent while DQN’s performance deteriorates.

In wind farms wind turbines are often placed close to each other. Each turbine generates a turbulent wake field, this field negatively affects subsequent turbines. This can cost more than 12% efficiency. To decrease this loss we can steer the turbines away from the wind direction, this will decrease the individual turbine power output, but can increase the total power output of the farm. As the size of the farm increases the number of possible actions increase exponentially. Due to this a numerical solution is not feasible. A reinforcement learning technique has been proven useful in the past, but a standard single agent implementation is still very computationally expensive. We evaluate the effectiveness of MADDPG on the active wake control problem. MADDPG is a multi agent reinforcement learning algorithm. MADDPG will be compared to the numerical solver FLORIS and to the already implemented and proven TD3 (which is a variation on a single agent DDPG algorithm). We compare the eventual output power of the algorithms with MADDPG. From the results we can see that MADDPG does improve on the learning performance of TD3, but since MADDPG needs to manage more neural networks the overhead is larger. MADDPG reaches an optimum solution in less training steps, but these steps take significantly more time.

Industry 4.0 and the Industrial Internet of Things (IIoT) growth will result in an explosion of data generated by connected devices. Adapting 5G and 6G technology could be the leading enabler of the broad possibilities of connecting IIoT devices in masses. However, the edge solution has some disadvantages, such as the loss of resource elasticity compared to cloud solutions. The research questions of this thesis are whether Deep Neural Networks (DNN)-porting can solve the accuracy-performance trade-off of edge computing solutions and how to implement an edge computing system based on open-source container-orchestrated DNN model inference platforms to enable vertical model autoscaling capabilities.
The thesis shows how porting techniques like structured pruning on DNN enable the accuracy performance trade-off in hardware-constrained settings. It generates models with reduced complexity and size while minimally degrading the accuracy. By using these ported models in the proposed inference platform, the thesis demonstrates how an edge computing system can achieve vertical model autoscaling capabilities, enabling efficient use of computational resources. This research focuses on CPU hardware and Real-Time (RT) request scenarios, where the latency Service Level Objective (SLO) combined with current demand are crucial factors. When the resources in an inference system deplete, the latency of individual requests can increase significantly due to queuing. The results show how an orchestrator can make live model version selections based on the model versions and demand. The proposed system increases the maximum possible throughput compared to the state-of-the-art while avoiding creating a queue in the RT scenario and improving system accuracy when CPU resources are available. Additionally, this work proposes a design to implement these benefits in industry-adopted open-source DNN inference platforms.

Artificial intelligence, machine learning, and deep learning have been the buzzwords in almost every industry (medical, automotive, defense, security, finance, etc.) for the last decade. As the market moves towards AI-based solutions, so does the computation need for these solutions increase and change with time. With the rise of smart cities and cyberphysical systems, the need for edge devices and efficient computation on the edge increases. While most of these newly developed deep learning models are quite large and wasteful in terms of energy, there have been recent methods that help improve the performance on the edge. However, due to their size, variety, and irregularity, the computing and power requirements are often too large to deploy these models on edge devices. This prohibits the application of such models within a rich field of application that requires high-throughput and real-time execution.

SENeCA (Scalable Energy Efficient Neuromorphic Computing Architecture) is a next-generation RISC-V-based neuromorphic computing architecture that was designed primarily for ultralow-edge applications where adaptivity is required. To mathematically model SENeCA, SENSIM (Scalable Energy Efficient Simulator, an open source simulator developed by the Interuniversity Microelectronic Center) provides an accurate mathematical software model of SENeCA, which helps in the early development and realization of a spiking neural network and deep neural network. This thesis work develops an efficient mapping tool SENMap (Scalable Energy-Efficient Neuromorphic Computing Architecture Mapper) on top of SENSIM which maps spiking neural networks efficiently. Having a faster, scalable realization software solution that can cater to large-scale neural networks can speed up the development procedure.

SENMap is developed in such a way that it supports flexible SNN/DNN application replacement, multiple single- and multi-objective optimization algorithms; the flexibility to choose from different optimization strategies; and also varying architectural parameters at the time of experimentation. Results show that mapping and neural processing elements (NPEs) depend primarily on the rate at which the sensor processes the data. On the basis of the rate, an early realization of SNN- and DNN-based edge AI chips SENMap. Depending on the actual parameters used, the maximum achieved improvements in energy consumption was around ~40%.

With the increase in the number of user devices and applications, the 5G systems (5GS) user plane is bound to be burdened. More work towards independent scaling and optimisation of the 5GS user plane has to be done. The 4G base stations (eNB) are mainly deployed as monolithic units, whereas, in 5GS, the 3GPP 5G Next Generation base station (gNodeB) can be divided into Radio Unit (RU), Distributed Unit (DU) and Centralised Unit (CU). The CU can be divided into two logical components, the Centralised Unit-Control Plane (CU-CP) and the Centralised Unit-User Plane (CU-UP), extending Control and User Plane Separation (CUPS) approach into RAN. With the introduction of CUPS into RAN, more independence will be provided to the user plane. Moreover, using Open RAN (O-RAN), all these components can be deployed at different locations as Virtual Network Functions (VNFs). We propose a 5GS architecture in this study, which optimises the user plane.

The key objectives of this thesis are twofold. Firstly, to compare the impact of co-locating the CU-UP with the DU at a distributed edge cloud location against co-locating the CU-UP with the CU-CP at a centralised regional cloud location. Secondly, with the aid of functional split design, virtualisation and O-RAN, we want to explore whether dynamically deploying the CU-UP and DU on a single physical host machine at a distributed location while centralising the CU-CP can enhance RAN development. To achieve this, we consider the 3GPP architecture as a reference and propose a new architecture and enhanced communication mechanism between CU-UP and DU. An analytical model was designed to evaluate the proposed architecture’s latency gains in the IP transport network, and a simulation model was designed to evaluate the proposed architecture’s communication latency. Furthermore, flow diagrams involving signalling of PDU session establishment are also presented.

We present an analysis and overall evaluation of the proposed architecture by comparing it with the reference architecture based on practical architectural aspects and PDU session signalling diagrams. The results of the calculation model and outputs of the simulation model indicated a significant latency improvement when the new architecture is employed. The new architecture found that, on average, 1.5 ms/packet of midhaul delay was reduced. And based on the flow diagram comparisons, it was found that the new architecture introduces overhead in terms of control plane signalling.

In this thesis we propose a novel memory architecture design that is robust to frequent memory failures targeting next generation low power embedded system. We explore the how the architecture works and perform detailed evaluations to show that our system achieves better performance than the state-of-the-art.

This work presents the first robust connection mode for Bluetooth Low Energy on an energy harvesting sensor.