Studentische Arbeiten

The goal of the thesis would be to build an Anticipatory or Proactive Wireless Resource Allocation Framework to optimize Multi-path networks.

The approach is to develop an optimization problem to allocate network resources to users by looking into window of time in the future and to solve this problem using Reinforcement Learning.

By knowing the channel quality of the users in the future, a better, more optimal allocation of resources is made possible.

Related Reading:

Chen, Weixi, et al. "Proactive 3C Resource Allocation for Wireless Virtual Reality Using Deep Reinforcement Learning." 2021 IEEE Global Communications Conference (GLOBECOM). IEEE, 2021

If you are interested in this work, please send me an email with a short introduction of yourself along with your CV and grade transcript.

• Strong Python programming skills
• Strong foundation on wireless communications
• Experience with Reinforcement Learning

Optical switches are an integral part of core networks. In this work, the student must model and evaluate the availability of the optical switch. The tasks are as follows:

• Identify subcomponents of the optical switch.
• Find the failure rates of these subcomponents based on data sheets.
• Model the subcomponents as individual Stochastic Activity Networks (using Mobius).
• Find the availability of the subcomponent and the switch.

The student is expected to use the Mobius software tool to model the Stochastic Activity Network. It is a simple tool that can be learned a week before the internship. 

Knowledge of any of the following is advantageous.

• Communication Network Reliability
• Analysis, Modelling, and Simulation of Networks
• Optical Networks

Software-Defined Networking (SDN) is a network paradigm where control and data planes are decoupled. The control plane consists on a controller, which manages network functionality and can be deployed in one or multiple servers. The data plane consists on forwarding devices which are instructed by the controller on how to forward traffic.

P4 is a domain-specific programming language, which can be used to define the functionality of forwarding devices as virtual or hardware switches and SmartNICs.

This work consists on performing measurements for a given P4 code on different devices. For that, an small P4-enabled virtual network will be used to perform some measurments. Later, data will be also collected from hardware devices as switchs and SmartNICs. Measurement should be depicted in a GUI for its subsequent analysis.

Basic knowledge on the following:

• Linux
• Networking/SDN
• Python/C
• Web programming (GUI).

Please send your CV and transcript of records.

Wireless communication is heavily dependent on the channel it operates on. Therefore, to simulate wireless transmissions, accurate wireless channel models are needed. The amount of channel models for aircraft cabins is quite limited and outdated because of the new materials used in aircrafts and new frequencies available for wireless transmission. Therefore, we want to develop a 3D structure of an two aisle aircraft cabin in Blender to simulate the propagation of electromagnetic waves. The goal of the thesis is to generate a wireless channel description of this model using ray-tracing and compare it to an existing single aisle model.

Review of related literature
Convert the existing single aisle 3D model to a double aisle
Derive a channel description
Compare the channel with an existing single aisle model
Evaluate the results

Implement a quantized neural network in P4 and evaluate the throughput of feed-forward networks and networks with attention mechanisms on P4 hardware.

Cloud Native deployments of the 5G Core network are gaining increasing interest and many providers are exploring these options. One of the key technologies that will be used to deploy these Networks, is Kubernetes (k8s).

In 5G, NG Application Protocol (NGAP) is used for the gNB-AMF (RAN-Core) communication. NGAP uses SCTP as a Transport Layer protocol. In order to load balance traffic coming from the gNB towards a resilient cluster of AMF instances, a L4 load balancer needs to be deployed in the Kubernetes Cluster.

The goal of this project is do develop a SCTP Load Balancer to be used in a 5G Core Network to aid the communication between the RAN and Core.
The project will be developed using the language Go (https://golang.org/).

- General knowledge about Mobile Networks (RAN & Core).
- Good knowledge of Cloud Orchestration tools like Kuberentes.
- Strong programming skills. Knowledge of Go (https://golang.org/) is a plus.

Software-Defined Radio Access Network (SD-RAN) is receiving a lot of attention in 5G networks, since it offers means for a more flexible and programmable mobile network architecture.

The heart of the SD-RAN architecture are the so called SD-RAN controllers. Currently, initial prototypes have been developed and used in commercial and academic testbeds. However, most of the solutions only contain a single SD-RAN controller. Nonetheless, a single controller becomes also a single point of failure for a system, not only due to potential controller failures but also due to a high load induced from the devices in the data plane.

To this end a multi-controller control plane often becomes a reasonable choice. However, a multi-controller control plane renders the communication among the controllers more challenging, since they need to often exchange control information with each other to keep an up to date network state. Unfortunately, currently there is no protocol available for such a communication.

The aim of this work is the development and implementation of an East/West API for SD-RAN controller communication according to 5G stardardization. The protocol should enable the exchange of infromation among the SD-RAN controllers regarding UEs, BSs, wireless channel state and allow for control plane migration among controllers.

• Experience with programming languages Python/C++.
• Experience with socket programming.
• Knowledge about SDN is a must.
• Knowledge about 4G/5G networks is a plus.

...

...

Das Thema dieser Bachelorthesis ist die Entwicklung, Optimierung und Evaluation des proprietären Brandmeldebussystems LSNi1 der Firma Bosch Sicherheitssysteme GmbH in Grasbrunn.

Brandmeldeanlagen sind ein bestehender notwendiger Teil der Gebaudetechnik und sollen fur neue vernetzte IoT-Dienste verwendet werden. Aufgrund immer höheren Anforderungen dieser Dienste soll die Datenrate uber den Brandmeldebus gesteigert werden. Zur Zeit wird das Bussystem zur verhaltnismäßig datenarmen Alarmabfrage und Antwort der Netzelemente verwendet. Ein wichtiger Punkt ist dabei, die Systemcharakteristika des Brandmeldesystems nicht zu verschlechtern. Solche Netzwerke sind als Sicherheitssysteme deklariert und mussen daher äußerst zuverlässig in Sachen Ausfallsicherheit und Fehlererkennung sein.

Die Aufgabe besteht darin, eine neue Übertragungstechnik mit höherer Datenrate auf der physikalischen Schicht des Busses zu finden und diese mit Hilfe eines Prototypen umzusetzen. Die Optimierung des Prototypen soll durch Berechnung und Experimentieren mit Systemelementen ausgearbeitet werden. Daraufhin werden verschiedene Testaufbauten, im firmeneigenen Labor getestet. Die dabei aufgenommenen Daten werden über ein entwickeltes Matlab-Evaluierungsprogramm ausgewertet, um eindeutige Aussagen über die Funktionalitat des Systems zu geben. Eine besondere Herausforderung bei Systemen dieser Art sind die hohen Kabellängen und die große Anzahl an Systemelementen auf dem Bus. Mit den ausgewerteten Ergebnissen soll überpruft werden, in welchem Umfang die Datenrate auf dem Bus gesteigert werden kann.

This thesis researches the optimal and joint path selection and adaptive function split distribution for automated vehicles in an industrial use case.

...

Conventional network performance analysis approaches consist of theoretical analysis, simulation studies, and testbed measurements.

Theoretical approaches, e.g. queuing theory, impose strict assumptions on network conditions for modeling.

However, these assumptions generally do not hold in real networks and therefore lead to unrealistic performance benchmarks.

An alternative to the theoretical approach is computational modeling, which is also referred to as network simulation studies in the literature.

The network simulators are implemented in software to include a wide range of network elements and protocols.

The simulation approach is cheap and faster to implement in comparison to a real testbed setup.

However, it is computationally expensive and they cannot simulate complex scenarios in real-time.

The final method of conventional testbed measurements aims to build a small-scale prototype to measure network performance.

The limitation of this approach is the cost and complexity of building a setup consisting of real hardware.

An alternative to conventional network performance evaluation is machine learning (ML) based approaches.

Recent research on ML applications to network performance evaluation focuses on learning graph neural network (GNN) models for analysis.

GNN-based modeling allows the expression of network topologies as graphs and their generalization capabilities allow prediction for larger topologies that are unseen during the training phase.

Popular examples of this are RouteNet and xNet.

The objective of this thesis is to implement RouteNet and xNet models to establish their performance benchmarks and compare them. 

The first step is to build a packet-level simulator for data collection and establish the ground truth for GNN models.

The next steps are to implement the models and evaluate them.

Hintergrund

Die Versorgung mit medizinischer Fachexpertise ist in vielen Gebieten
der Welt aufgrund einer Zentralisierung der Gesundheitssysteme sowie
mangelhafter Infrastruktur eingeschränkt. Zudem besteht ein Mangel
an spezialisierten Ärzt:innen-und Pflegekräften. Ziel sollte der Aus-
gleich der Dysbalance zwischen Ballungsräumen und ländlichen Gebie-
ten sein, wozu mobile Datenübertragungstechnologien einen wichtigen
Beitrag leisten können. 6G bietet die Voraussetzungen, medizinische
Telepräsenz und hochpräzise Telediagnostik zuermöglichen. Im Pro-
jekt 6G-life soll unter anderem ein Demonstrator für ein robotisches
Teleoperationssystem im 6G-Testbed entwickelt werden, welches für
eine diagnostische Auskultationgerüstet und einer klassischen, per-
sönlichen Untersuchung gleichgestellt ist.

Aufgabenstellung

Ziel der Arbeit ist die Entwicklung einer containerbasierten Steuerung
für ein robotisches Teleoperationssystem unter Nutzung von fusionier-
ten Sensordaten. Hierfür soll zunächst eine ausführliche Literatur-
recherche zugängigen Steuerungskonzepten in ROS sowie zur Fusion
von Sensordatendurchgeführt werden.Vorhandene Lösungsansätze
sollen im nächsten Schritt bewertet und verglichen werden. Darauf ba-
sierend sollen Funktionskonzeptefür eine containerbasierte Steue-
rungsarchitekturerarbeitet und unter Einbindung mehrerer Kamera-
sensoren am Demonstrator implementiert werden. Abschließend soll
das ausgearbeitete System mittels geeigneter Funktionsversuche hin-
sichtlich Flexibilität/Latenz/Kommunikationsfähigkeit evaluiert wer-
den.

Teilaufgaben

Im Rahmen der Arbeit sollen folgende Punkte behandelt werden:
• Literaturrecherche zuROS,Cloud Robotics,Sensordatenfusion
• Bewertung und Vergleich vorhandener Lösungsansätze
• Erarbeitung von Konzeptenfür eine containerbasierte ROS-Ar-
chitekturmit Integration mehrerer Kamerasensoren
• Implementierung der Konzepte am Demonstrator
• Design, Durchführung und Evaluation von Funktionstestshin-
sichtlich Flexibilität/Latenz/Kommunikationsfähigkeit

Time-Sensitive Networking (TSN)is a set of standards[1]developedby IEEE 802.1 Task Groupto enableEthernet networks to giveQuality of Service (QoS)guarantees for time-sensitiveor mission-critical traffic and applications.VariousTSN standards provide differing QoSguaranteesand require different functions to be implemented in hardware. As devices
frommultiplevendorsneedtooffermutuallycompatiblefunctions,profilessuchas IEEE60802forIndustrialAutomation[2]arebeingdefined.These profilesfocusona commonset of functions and configurations in order to decreasethe complexity which possible variations in standards might create.

REST API, Knowledge and Experience with 5G systems, Undesrstanding of TSN.

Python.

Cell-free massive MIMO (cfmMIMO) is a promising system that removes the restriction in traditional MIMO networks that an access point (AP) can only serve user elements (UE) within its immediate vicinity – i.e. the cell. Instead, a large number of multi-antenna APs distributed over a service area is used to coherently serve UEs in the same time/frequency resource [MLYN16]. [NAYL+17] has shown that cell-free massive MIMO has significantly better performance than conventional small-cell systems which limit each UE to be served by one AP only. Despite increased spectral efficiency, some have questioned the energy efficiency of cfmMIMO systems that might affect the overall practicability of cfmMIMO, especially as such systems require a large number of backhaul links which potentially increase the total power consumption to such a level that can overwhelm the spectral efficiency gains [NTDM+18].

In light of the increasingly active discussions around cfmMIMO energy efficiency, we consider such a system in the uplink configuration for simplicity and assume each AP has a fixed power consumption. We propose to minimize the sum power consumption of all APs with respect to active APs, UE transmit power, and beamformers, subject to UE maximum transmit power constraint and minimum per-UE achieved ergodic rate. For a detailed mathematical description of the optimization problem, please see the attached PDF file.

The rationale behind minimizing AP power consumption instead of UE transmitted/radiated power is because the per AP power consumption is in the order of kilowatts, while UEs transmit in the order of milliwatts. The power saving potential is simply much larger by optimizing the former.  

The student will attempt to solve this combinatorial problem using advanced machine learning methods, specifically looking into Bayesian optimization due to the combinatorial nature of the problem hindering the application of gradient-based methods. Additionally, to preserve scalability and avoid imposing a Gaussian prior onto the objective function, the usage of deep ensembles [LPB17] instead of a Gaussian process could be more appropriate.

Emphasis will be placed on real-time optimization, i.e. to arrive at a decision configuration quickly enough to adapt to changing network behavior. There will also be an option to model the AP's power consumption more realistically and introduce more complexity to the cost function. After optimization, a detailed analysis will be carried out on the results, such as a comparison of the performance of the proposed algorithm on different channel modelling conditions to evaluate the practicability of the algorithm in real-world scenarios.

[MLYN16]  T. L. Marzetta, E. G. Larsson, H. Yang, and H. Q. Ngo, Fundamentals of Massive MIMO. Cambridge, U.K.: Cambridge Univ. Press, 2016.

[NAYL+17] H. Q. Ngo, A. Ashikhmin, H. Yang, E. G. Larsson, and T. L. Marzetta, “Cell-free massive MIMO versus small cells,” IEEE Trans. Wireless Commun., vol. 16, no. 3, pp. 1834– 1850, Mar. 2017.

[NTDM+18] H. Q. Ngo, L.-N. Tran, T. Q. Duong, M. Matthaiou, und E. G. Larsson, „On the Total Energy Efficiency of Cell-Free Massive MIMO“,IEEE Transactions on Green Communications and Networking, Bd. 2, Nr. 1, S. 25–39, Ma?rz 2018, doi: 10.1109/TGCN.2017.2770215.

[LPB17] B. Lakshminarayanan, A. Pritzel, und C. Blundell, „Simple and Scalable Predictive Uncertainty Estimation using Deep Ensembles“. arXiv, 3. November 2017 [Online]. Verfu?gbar unter: http://arxiv.org/abs/1612.01474. [Zugegriffen: 24. Ma?rz 2023]

Indoor positioning is a crucial feature for diverse industrial use cases, as it allows for improved monitoring and task automation, thus enhancing production efficiency. Nonetheless, inexpensive positioning deployments based on signal strength measurements struggle to provide good localization accuracy, as these measurements are usually highly variant.

Fingerprinting based on received signal strength indicators (RSSI) has been proposed as a promising approach to mitigate the high variability and lead to relatively high positioning accuracy while keeping deployment costs low. Nonetheless, using RSSI fingerprinting for accurate positioning entails mapping large and noisy vectors of RSSI measurements to specific points in space, which is a challenging task.

Several techniques have been evaluated in the state of the art to perform this mapping: distance estimation followed by trilateration, k-nearest neighbors averaging, radiomap interpolation and error minimization, and, in recent years, also neural networks. By training neural networks with sufficient amounts of accurate data, we can abstract away the complexity of the mapping and still achieve accurate positioning.

Nonetheless, as in other applications, it has been observed that neural networks often struggle at generalizing data beyond those close to the training set, and their complete lack of knowledge about the underlying physical phenomena may result in obviously wrong results that are difficult to prevent with simply more training.

Motivated by these and other related facts, in recent years the concept of graph-network- based learning has emerged. As opposed to neural networks, graph networks use graphs to represent all inputs and outputs of the learning process and apply modified versions of common training approaches to convert input graphs into output graphs. The intention behind this procedure is that graphs themselves can be defined in such a way to model our a priori knowledge about the inputs and outputs.

In this thesis, we will investigate the application of graph-network-based machine learning to RSSI fingerprinting for localization, with the intention of incorporating radio propagation and environment models into the learning process. The results will be compared against other state-of-the-art ML approaches to conclude whether graph networks can help in producing more accurate positioning from the same training data.

• Python/C++/Matlab
• Machine Learning basics
• Wireless communication basics
• Mmathematical and analytical skills

In this thesis, we want to build a decentralized FL on embedded systems. Unlike traditional FL, the model is only shared between the local microcontrollers without a central server. Each microcontroller generates its global model by communicating with the neighbor nodes in a self-organized network. The purpose is to evaluate the training performance while enhancing communication efficiency to achieve the limitation of this exclusive microcontroller platform.

Wireless communication, experience in Machine Learning, Python. 

Redundancy for reliability is an important aspect of deterministic networking. A legacy protocol called Media Redundancy protocol and a newer proposal called Frame Replication and Elimination for Reliability are good candidates in this regard.

This thesis will give an overview of all current and relevant redundancy protocols. It will analyze MRP and FRER in-depth to find the needs of these protocols to enable interoperability. Furthermore, a novel concept of network components to solve the interoperability problems re- regarding redundancy for Industrial Ethernet is presented. It is expected that a component must be configured which enables devices with the two different redundancy protocols implemented to communicate. Tests are performed on a testbed to prove the applicability of the theoretical concept. An evaluation of different topologies will also be made. Finally, the metrics of MRP, FRER, and MRP-FRER networks are compared to identify the performance differences that may arise and which configuration still fulfills the real-time requirements, adds benefits to the user, and therefore can be recommended for Ethernet-based communication in automation and control systems.

C and C++, Knowledge of Routing and Forwarding, Ethernet systems. Experience with TSN is a plus.

The Radio Access Network (RAN) architecture for cellular networks is developing towards exposing an increasing number of data collection and control interfaces. The applications I will develop in my master’s thesis are hosted on a platform named EdgeRIC, which operates as a real-time RAN controller sitting next to the Distributed Units (DU) inside a 5G cellular network architecture. The proximity of the EdgeRIC platform to the transmitting and receiving base stations allows for real-time low latencies when processing information from the RAN. My applications are examples of how RAN control can benefit from this real-time information.

1. Accelerated Digital Twin of the Network

This application will host an emulated copy of a real-world physical RAN. The copy will be an instance of a srsRAN network using computer networking transport channels instead of radio links. The goal is to accelerate the events inside the network copy to more frequently generate data on metrics such as channel quality. When this can be achieved, a MAC scheduling policy using reinforcement learning, for example, could train on this feedback faster than in the real world and the trained policy could be applied to real-world scheduling decisions more quickly.

2. Mitigation of Interference from Neighboring Cells

This application tackles the problem of interference in frequency from neighboring cells by applying a beamformed null signal in the direction of the interfering base station. This should improve throughput for users attached to the original base station. The EdgeRIC platform ensures low latency for computing the null signal.

The goal of the thesis would be to build an Anticipatory or Proactive Wireless Resource Allocation Framework to optimize Multi-hop, Multi-path networks.

The approach is to develop an optimization problem to allocate network resources to users by looking into window of time in the future and to solve this problem using Reinforcement Learning.

By knowing the channel quality of the users in the future, a better, more optimal allocation of resources is made possible.

Related Reading:

Dastgheib, Mohammad Amir, et al. "Mobility-aware resource allocation in VLC networks using T-step look-ahead policy." Journal of Lightwave Technology 36.23 (2018): 5358-5370.

If you are interested in this work, please send me an email with a short introduction of yourself along with your CV and grade transcript.

• Strong Python programming skills
• Strong foundation on wireless communications
• Experience with Reinforcement Learning

Cloud computing and microservice-based architectures have empowered businesses to develop new highly reliable applications, that can adapt to variable workloads. In the context of 5G telco applications, both the research community and the industry has been exploring methods of using cluster orchestrators, such as Kubernetes (K8s) in mobile network deployments. More specifically, Multi-access Edge Computing and Fog computing for 5G networks represent use-cases where the principles of cloud computing can be applied, but meeting the requirements (especially regarding latency) proves challenging.

With the increased usage of cloud deployments for Radio Access Networks, cluster configuration gained importance, as an optimized configuration translates into higher performance, increased agility and better usage of the resources. Empirical, experience-based human heuristics can improve the cluster configuration, however they require advanced knowledge about the deployment and the direct intervention of the cluster operator.

The research community is currently exploring the steps towards an automated, data-driven cluster configuration: the behavior of the cluster is learned with Machine Learning (ML), the cluster behavior is simulated with different configurations and an optimizer chooses the best configuration. However, an optimised configuration is only applicable to the real-life cluster, if the simulation of the cluster behavior is highly accurate.

The goal of this Master's Thesis is to determine the net value added by building a Digital Twin of a k8s cluster. Therefore, it uses ML-based models for the simulation of three network functions compared to traditional “Hand-crafted models”. First, it implements new network functions such as the pod scheduler and the load balancer in an existing simulation framework for k8s cluster.

Second, the thesis compares classical, hand-crafted models for the simulation with data-driven methods. Namely, after implementing Load Balancing and Pod Scheduling in the simulator in the classical way, it also integrates already trained ML model equivalents of these functions.
In the end, performance metrics and accuracy of both appraoches are compared.

Bayerischer Rundfunk (BR) operates a network to deliver content via television, radio and the internet to its users. This requires a highly heterogeneous network. The network monitoring solution for the BR-network collects alert data from involved devices and stores it in a central database. Currently, human operators make network management decisions based on a manual review of this data. This especially includes root cause identification in case of network failures. Such a human-centric process can be tedious and does not scale well with increasing network complexity. In this thesis, the student should perform a thorough analysis of the described data and characterize service disruptions to enable automated analysis in the future.

Mobile Edge Computing (MEC) enabled 6G network support low latency applications running in energy-constrained and computational limited devices, especially IoT devices. Using the task offloading concept, the devices offload the incoming tasks fully or partially to MEC depending on the device and network side's communication and computation resource availability.

The 6G networks are oriented towards the Digital Twins (DT); therefore, the resource allocation and offloading decisions are enhanced with the context-awareness of the devices, environment, and network. The device context awareness consists of battery state, power consumption, CPU load, and traffic type. Further, the environmental context-awareness includes the position of the network components, the mobility patterns, and the predicted routes. Even though most IoT devices are accounted for as static devices, there are use cases in NR-Light where devices obtain mobility.

In this thesis, the student will focus on developing and testing a context-aware communication and computing resource allocation mechanism, focusing on decreasing individual devices' energy consumption and reducing processing latency.

• Good knowledge of Python and Matlab programming.
• Good mathematical background.
• Knowledge of mobile networks.

Network Intrusion Detection (NID) is a common topic in cybersecurity. However, it is not trivial to find a solution when facing the complicated network environment nowadays. Often a complex system is needed to process enormous volume of data stored in databases. This thesis proposes to use Deep Learning models to tackle the NID problem in a pre-train/fine-tune manner. As the new paradigm of transfer learning, the process of pre-training follows by fine-tuning has achieved huge success in many areas such as vision and NLP. We aim to study whether those trending models still perform well on large-scale structured data such as network security logs. It is plausible to leverage the strong learning ability of DL models to learn table representations and separate anomaly from benign records based on the learned information.

• Machine learning knowdlege
• Programming skills (Python, GIT)
• Computer networking knowledge

Aeronautical applications such as satellite communications and in aircraft systems have enhanced the range of applications for 5G, bringing new opportunities and potentials.

In the new 6G era these industries are envisioned to receive even more attention due to the high coverage possibilities that they provide. Applications such as unmanned aerival vehicles, flying taxis, moving base stations are just a few to name. In that regard, to support these wide range of applications, new 6G network urge for new and more efficient architectures.

In this thesis the student shall focus on the analysis of existing archiectures for 5G/6G for aeronautical applications and shall perform a comparison, definining potentials for the development of new architectures to reduce delay and increase the network performance. To this end, a comprehensive analysis is required to define the right metrics for comparison and identifying the potentials for improvement. An initial simulator based on the defined metrics is to be created to enable the comparison of similar algorithms in the future.

    Good knowledge of simulation environments such as Matlab, Python.
    Good mathematical background.
    Knowledge about satellite communications is a plus.

In order to fully utilize the capabilities of a LiFi-RF Heterogeneous network, the client devices should be capable of using multiple network interfaces simultaneously. Thanks to multipath solutions like MPTCP, this is possible. 

The challenge in a MPTCP-enabled heterogeneous network lies in designing a policy to schedule data packets onto the multiple paths with heterogeneous characteristics (eg. delay, packet loss). 

This work involves

• Re-implementing an existing deep reinforcement learning model of a multipath scheduler
• Extending the scheduler
• Evaluating the models extensively in an emulation environment (Mininet)
• Evaluating the models extensively on the hardware testbed

If you are interested in this work, please send an email with a short introduction of yourself along with your CV and grade transcript.

• Strong Python programming skills
• Experience with reinforcement learning
• Experience with Linux
• Experience using Mininet is an advantage

Today's Data Center (DC) networks are facing increasing demands and a plethora of requirements. Factors for this are the rise of Cloud Computing, Virtualization and emerging high data rate applications such as distributed Machine Learning frameworks.
Many proposal for network designs and routing algorithms covering different operational goals and requirements have been proposed.
This variety makes it hard for operators to choose the ``right'' solution.
Recently, some works proposed that automatically generate adversarial input to networks or networking algorithms [1,2] to identify weak spots in order to get a better view of their performance and help operators' decision making. However, they focus on specific scenarios.
The goal of this thesis is to develop or extend such mechanisms so that they can be applied a wider range of scenarios than previously.
The thesis builds upon an existing flow-level simulator in C++ and initial algorithms that generate adversarial inputs for networking problems.

[1] S. Lettner and A. Blenk, “Adversarial Network Algorithm Benchmarking,” in Proceedings of the 15th International Conference on emerging Networking EXperiments and Technologies, Orlando FL USA, Dec. 2019, pp. 31–33, doi: 10.1145/3360468.3366779.
[2] J. Zerwas et al., “NetBOA: Self-Driving Network Benchmarking,” in Proceedings of the 2019 Workshop on Network Meets AI & ML  - NetAI’19, Beijing, China, 2019, pp. 8–14, doi: 10.1145/3341216.3342207.

- Profound knowledge in C++

Bayerischer Rundfunk (BR) operates a network to deliver content via television, radio and the internet to its users. This requires a highly heterogenous network. The network monitoring solution for the BR-network collects log data from involved devices and stores it in a central database. Currently, human operators make network management decisions based on a manual review of this log data. This especially includes root cause identification in case of network failures. Such a human-centric process can be tedious and does not scale well with increasing network complexity. In this thesis, the student should perform a thourough analysis of the described data and evaluate the potential for automated processing. Goal is to provide a data-driven approach that significantly supports human operators with identifying root causes in case of network failures.

Software Defined Networking (SDN) separates the control and data planes.Control plane can be considered as the brain of the network and it is responsible for configuring flows, finding paths and managing all the network functionalities like firewall,  load balancing,  etc.  For this reason,  the SDN controller became complex.  Furthermore, it is a large software platform, which have many contributors  with  different  experience  level.   As  a  result  the  code  contains  many undetected and unresolved bugs.  If one of these bugs is activated in the operational state, it may cause performance degradation or even collapse of the whole system.

SDN serves to broad range of applications with different requirements.  Some of the application areas like autonomous driving requires high reliability and performance degradation may cause undesired results.  Software Reliability Growth Models (SRGM) are statistical frameworks that are based on historical bug reports  for  reliability  analysis  and  widely  used  to  estimate  the  reliability  of  a software.  Open network operating system (ONOS) is an open source project and it became one of the most popular SDN platforms.  Its historical bug reports are open in their JIRA issue tracker.  Currently ONOS has 23 releases, its first ten  versions  are  investigated  with  different  SRGM  models  [1]  and  found  that different SRGMs fit to the bug detection of different versions of ONOS.

Source code metrics refer to quantitative characteristics of the code.  Those metrics  can  describe  the  size  of  the  code  (lines  of  code),  complexity  of  code (McCabe’s complexity), etc.  They have been used to predicting the number of bugs, identifying possible potential location of bug, etc.

The goal of this work is to analyse the reliability of different ONOS releases. For that purpose, an understanding of the correlation between the structure of source code and the bug manifestation process is crucial to predict the future bug manifestation of the new releases.  First, a state of the art research on the SRGM will  be  done  to  understand  the  software  reliability  and  SRGMs.   Afterwards the  student  should  implement  different  SRGMs  to  fit  the  error  manifestation of  every  release  and  compare  the  results  with  mentioned  research  [1].   Then, different  code  metrics  will  be  obtained  from  each  ONOS  release.   Then,  the correlation between SRGM and code metrics will be revealed.  At last reliability of the release will be analyzed with the best fitting SRGM. The result of this work will be to propose a reliability metric combining SRGM and code metrics that improves the software reliability prediction.

References

P. Vizarreta, K. Trivedi, B. Helvik, P. Heegaard, W. Kellerer, and C. Mas Machuca, An empirical study of software reliability in SDN controllers,  13th International  Conference  on  Network  and  Service  Management  (CNSM), 2017.

In the Radio Access Network (RAN), the MAC scheduler is largely inherited across generations in the past, to fit to new networking goals and service requirements. The rapid deployment of new 5G technologies will make upgrading of current ones extremely complicated and difficult to improve and maintain. Therefore, finding new solutions for efficient Radio Resource Scheduling (RRS) is necessary to meet the new KPI targets. 5G networks and beyond use the concept of network slicing by forging virtual instances (slices) of its physical infrastructure. A heterogeneous network requires a more optimized and dynamic RRS approach. In view of the development of SD-RAN controllers and artificial intelligence, new promising tools such as reinforcement learning can be proven useful for such a problem.

In this thesis, a data-driven MAC slice scheduler will be implemented, that maximizes user utility, while learning the optimal slice partitioning ratio. A deep reinforcement learning technique will be used to evaluate the radio resource scheduling and slicing in RAN. The results will be compared with traditional schedulers from the state-of-the-art.

...

In this thesis student will focus on designing and implementing a hierarchical SDN solution for industrial multi-domain TSN network.

Recently, a scalable load balancing algorithm in the dataplane has been proposed that leverages P4 to estimate the utilization in the network and assign flows to the least utilized path. This approach can be interpreted as a form of Graham's List algorithm.

In this thesis, the student is tasked to investigate how a different online scheduling algorithm called IMBAL performs compared to HULA. A prototype of IMBAL should be implemented in NS3. The tasks of this thesis are:

1. Literature research and overview to online scheduling and traffic engineering in data center networks.
2. Design how IMBAL can be implemented in NS3.
3. Implementation of IMBAL in NS3.
4. Evaluation of the implementation in NS3 with production traffic traces and comparison to HULA (a HULA implementation is provided from the chair and its implementation not part of this thesis).

Next generation mobile networks are envisioned to cope with heterogeneous applications with diverse requirements. To this end, 5G is paving the way towards more scalable and higher performing deployments. This leads to a revised architecture, where the majority of the functionalities are implemented as network functions, which could be scaled up/down depending on the application requirements. 

3GPP has already released the 5G architecture overview, however there exists no actual open source deployment of RAN functionalities. This will be crucial towards the evaluation of the Core network both in terms of scalability and performance. In this thesis, the student shall understand the 5G standardization, especially the control plane communication between the RAN and 5G Core. Further, an initial RAN function compatible with the 5G standards shall be implemented and evaluation of control plane performance will be carried out. 

• Strong knowledge on programming languages Python, C++ or Java.
• Knowledge about mobile networking is necessary.
• Knowlegde about 4G/5G architecture is a plus.

In order to fully utilize the capabilities of a LiFi-RF Heterogeneous network, the client devices should be capable of using multiple network interfaces simultaneously. Thanks to multipath solutions like MPTCP, this is possible. 

The challenge in a MPTCP-enabled heterogeneous network lies in designing a policy to schedule data packets onto the multiple paths with heterogeneous characteristics (eg. delay, packet loss).

This work involves

• Designing an MPTCP scheduler that schedules packets optimally to minimize network delay and handle the dynamicity of heterogeneous links
• Implementing the scheduler in the Linux kernel
• Performing extensive evaluations with Mininet and hardware

Related Reading:Yang, Wenjun, et al. "Loss-aware throughput estimation scheduler for multi-path TCP in heterogeneous wireless networks." IEEE Transactions on Wireless Communications 20.5 (2021): 3336-3349.

If you are interested in this work, please send an email with a short introduction of yourself along with your CV and grade transcript.

• Strong Python and C++ programming skills
• Experience with optimization problems
• Experience with Linux networking

Mobile Edge Computing (MEC) enabled 6G network support low latency applications running in energy-constrained and computational limited devices, especially IoT devices. Using the task offloading concept, the devices offload the incoming tasks fully or partially to MEC depending on the device and network side's communication and computation resource availability.

The 6G networks are oriented towards the Digital Twins (DT); therefore, the resource allocation and offloading decisions are enhanced with the context-awareness of the devices, environment, and network. The device context awareness consists of battery state, power consumption, CPU load, and traffic type. Further, the environmental context-awareness includes the position of the network components, the mobility patterns, and the quality of the wireless channel and the availability of the network resources.

In this project, the student will focus on developing and testing an artificial intelligence algorithm for joint allocating of computing and radio resources in a predictive manner, focusing on decreasing individual devices' energy consumption and reducing processing latency.

Tasks

• Work with a 6G radio access network simulator, to generate the database for the scenario with devices having high energy efficiency and low task processing latency requirements.
• Develop a reinforcement learning algorithm for joint allocation of radio and computing resource allocation.
• Comparing the developed model with the state-of-the-art approaches.
• Test and documentation.

• Good knowledge of Python programming.
• Good mathematical background.
• Good knowledge of deep learning/reinforcement learning.

This work focuses on the exploration of networks enabling train control and on-board data communications under mobility scenarios. Today, low bandwidth networks such as GSM, providing less than 200 Kbps are being used to transmit train control information. Moreover, despite trains may use multiple on-board technologies to provide users with an internet connection (e.g., repeaters, access points), they fail in their attempt as these connections are characterized by having low throughput (less than 2 Mbps) and frequent service interruptions.

This work aims at the development of a network planning solution enabling future applications in train mobility scenarios such as: Automatic Train Operation (ATO) [1,2], leveraging cloud technologies and meeting bandwidth requirements of data-hungry end-users' applications. Here, special attention will be given to the migration of communications services triggered by trains mobility patterns. It is expected of the student to find solutions to the following questions:

• When to trigger service migrations?

• Where to migrate services? (i.e., to which data center)

• How to handle this process? (So that the user does not perceive any interruption)

 Given:

• Trains mobility patterns

• Service requirements in terms of bandwidth and delay

• Network topology

• Data center locations

 
The results from this work can be useful to get an insight on requirements for Smart Transportation Systems, that may in turn be useful for cementing the basis of other scenarios such as: Autonomous Driving and Tele-Operated Driving.

 [1] Digitale Schiene Deutschland. Last visit on 13.12.2021 https://digitale-schiene-deutschland.de/FRMCS-5G-Datenkommunikation

[2] 5G-Rail FRMCS. Last visit on 13.12.2021 https://5grail.eu/frmcs/

Basic knowledge in:

• Integer Linear Programming (ILP), heuristics or Machine Learning (ML).

• Python

Please send your CV and transcript of records.

The goal of the thesis would be to build an Anticipatory or Proactive Wireless Resource Allocation Framework to optimize Multi-path networks.

The approach is to develop an optimization problem to allocate network resources to users by looking into window of time in the future and to solve this problem using Reinforcement Learning.

By knowing the channel quality of the users in the future, a better, more optimal allocation of resources is made possible.

Related Reading:

Chen, Weixi, et al. "Proactive 3C Resource Allocation for Wireless Virtual Reality Using Deep Reinforcement Learning." 2021 IEEE Global Communications Conference (GLOBECOM). IEEE, 2021

If you are interested in this work, please send me an email with a short introduction of yourself along with your CV and grade transcript.

• Strong Python programming skills
• Strong foundation on wireless communications
• Experience with Reinforcement Learning

This work focuses on the exploration of networks enabling train control and on-board data communications under mobility scenarios. Today, low bandwidth networks such as GSM, providing less than 200 Kbps are being used to transmit train control information. Moreover, despite trains may use multiple on-board technologies to provide users with an internet connection (e.g., repeaters, access points), they fail in their attempt as these connections are characterized by having low throughput (less than 2 Mbps) and frequent service interruptions.

This work aims at the development of a network planning solution enabling future applications in train mobility scenarios such as: Automatic Train Operation (ATO) [1,2], leveraging cloud technologies and meeting bandwidth requirements of data-hungry end-users' applications. Here, special attention will be given to the migration of communications services triggered by trains mobility patterns. It is expected of the student to find solutions to the following questions:

• When to trigger service migrations?

• Where to migrate services? (i.e., to which data center)

• How to handle this process? (So that the user does not perceive any interruption)

 Given:

• Trains mobility patterns

• Service requirements in terms of bandwidth and delay

• Network topology

• Data center locations

 
The results from this work can be useful to get an insight on requirements for Smart Transportation Systems, that may in turn be useful for cementing the basis of other scenarios such as: Autonomous Driving and Tele-Operated Driving.

 [1] Digitale Schiene Deutschland. Last visit on 13.12.2021 https://digitale-schiene-deutschland.de/FRMCS-5G-Datenkommunikation

[2] 5G-Rail FRMCS. Last visit on 13.12.2021 https://5grail.eu/frmcs/

Basic knowledge in:

• Integer Linear Programming (ILP), heuristics or Machine Learning (ML).

• Python

Please send your CV and transcript of records.

...

....

...

Energy efficiency is one of the key performance requirements in the 5G network to ensure user experience. A portion of devices, especially the Industrial Internet of Things (IIoT), run on limited energy, supported by the batteries not placed over the lifetime.

Therefore, the estimation of the power consumption and battery lifetime has recently received increased attention. Multiple context parameters, such as mobility and traffic arrivals, impact the device's power consumption.

In this thesis, the student shall focus on analysing the impact of different traffic models on the power consumption of user equipment. Different source and aggregated traffic models will be implemented depending on the number of devices n the scenario. The implemented traffic models will be evaluated based on a context-aware power consumption model for the user equipment.

• Good knowledge of Python and Matlab programming.
• Good mathematical background.
• Knowledge mobile networks.

Increased interference is one of the main drawbacks of cell densification, which is an important strategy for 5G networks to achieve higher data rates. Function centralization has been proposed as a strategy to counter this problem, by letting the physical or scheduling functions coordinate among one another. Nevertheless, the capacity of the fronthaul network limits the feasibility of this strategy, as the throughput required to connect low level functions is very high. Fortunately, since not every function benefits in the same way from centralization, a more flexible approach can be used. Instead of centralizing all functions, only those providing the highest amount of interference mitigation can be centralized. In addition, the centralization level, or functional split, can be change during runtime according to the instantaneous network conditions. Nonetheless, it is not fully know how costly it is to deploy and operate a network implementing a dynamic functional split.

In this internship, the cost of a radio access network implementing a dynamic functional split will be evaluated. A simulator already developed at LKN will be used and extended to produce network configurations adapted to the instantaneous user position and activity. Then, off-the-shelf cost models will be improved and used to estimate the deployment and operating cost of the network under multiple scenarios. Furthermore, the conditions on which a dynamic functional split is profitable will be investigated. Improvements on the functional-split selection algorithm will be proposed, such that the operator benefits from enhanced performance without operating at exceedingly costly states. Finally, a model that takes into account the cost of finding and implementing a new functional split will be employed and its results compared to the previous results.

...

In electronics and telecommunication, jitter is a significant and an undesired factor. The effect of jitter on the signal depends on the nature of the jitter. It is important to sample jitter and noise sources when the clock frequency is especially prone to jitter or when one is debugging failure sources in the transmission of high speed serial signals. Managing jitter is of utmost importance and the methods of jitter decomposition have changed comparably over the past years.

In a system, jitter has many contributions and it is not an easy job to identify the contributors. It is difficult to get Random Jitter on a spectrogram. The waveforms are initially constant, but the 1/f noise and flicker noise cause a lot of disturbance when it comes to output measurement at particular frequencies in a system. 

The task is to understand the difference between the jitter calculations based on a step response estimation and the dual dirac model by comparing the jitter algorithms between the R&S oscilloscope and other competition oscilloscopes. Also to understand how well the jitter decomposition and identification is there.

The tasks in detail are as follows.

Setup a waveform simulation environment and extend to elaborate test cases

Run the generated waveforms through the algorithms

Analyze and compare the results:

Frequency domain

Statistically (histogram, etc)

Time domain

Consistency of the results

Evaluate the estimation of the BER (bit error rate)

Identify the limitations of the dual-dirac model

Compare dual-dirac model results with a calculation based on the step response estimation

Generate new waveforms based on the analysis

Summarize findings

The goal of this thesis is the classification of packet-level traces using Markov- and Hidden Markov Model. The scenario is open-world: Traffic of specific web applications should be distinguished from all possible web-pages (background traffic). In addition, several pages should be differentiated. Examples include: Google Maps, Youtube, Google Search, Facebook, Google Drive, Instagram, Amazon Store, Amazon Prime Video, etc.

...

...

...

The architecture of 5G radio access networks features the division of the base station (gNodeB) into a centralized unit (CU) and a distributed unit (DU). This division enables cost reduction and better user experience via enhanced interference mitigation. Recent research proposes the posibility to modify this functional split dynamically, that is, to lively change the functions that run on the CU and DU. This has interesting implications at the network operation.

In this topic, the student will employ a dedicated simulator developed by LKN to characterize the duration and transition rates of each functional split under multiple variables: population density, mitigation capabilities, mobility, etc. This characterization may be used then on traffic models to predict the network behavior.

MATLAB, some experience with mobile networks and simulators

The Wireless Sensor Networks lab offers the opportunity to develop software solutions for the wireless sensor networking system, targeting innovative applications. For the next semester, a position is available to assist the participants in learning the programming environment and during the project development phase. The lab is planned to be held on-site every Tuesday 15:00 to 17:00.

• Solid knowledge in Wireless Communication: PHY, MAC, and network layers.
• Solid programming skills: C/C++.
• Linux knowledge.
• Experience with embedded systems and microcontroller programming knowledge is preferable.

The main responsibilities of a working student include assistance to tutors in the correction of the programming assignments, as well as answering the questions in Moodle.  Working time is 6-7 hours per week in the period from May to July.

• Python knowledge

The results expected from this work are the enhancement of the 5G/6G tested setup with additional features on the Radio Access Network (RAN) and Core Network (CN). The work is focused on the OpenAirInterface (OAI) [1] platform, which forms the basis of the testbed setup. The expected outcome is to have improvements in wireless resource scheduling, focused on the uplink (UL), power management, and core network function management. 

[1] N.Nikaein, M.K. Marina, S. Manickam, A.Dawson, R. Knopp and C.Bonnet, “OpenAirInterface: A flexible platform for 5G research,” ACM SIGCOMM Computer Communication Review, vol. 44, no. 5, 2014.

• Good C/C++ experience
• Good Python knowledge
• RAN and CN architecture understanding is a plus

Availability is the probability that a device performs its required function at a particular instant of time.

In most networks, the components are brought from different manufacturers. They have different availabilities. Network operators prefer having reliable components handling more traffic. This ensures the robustness of the network. So, assigning appropriate manufacturers to the components in the topology to guarantee maximum availability is essential.

In this work, the student uses centrality metrics to identify the critical nodes and assign manufacturers based on these metrics.

Mandatory:

• Kommunikationsnetze course at LKN
• Python

Availability is the probability that a device performs its required function at a particular instant of time.

In most networks, the components are brought from different manufacturers. They have different availabilities. Network operators prefer having reliable components handling more traffic. This ensures the robustness of the network. So, assigning appropriate manufacturers to the components in the topology guaranteeing 
a) maximum availability, and 
b) load balancing on the nodes
is essential.

For a fixed topology and known traffic, how can the components be assigned to manufacturers to maximise availability and balance load on nodes?

Mandatory:

• Communication Network Reliability course/ Optical Networks course at LKN
• Python

Preferred:

• Knowledge of Linear Programming and/or nonlinear programming

Future medical applications put stringent requirements on the underlying communication networks in terms of highest availability, maximal throughput, minimal latency, etc. Thus, in the context of the 6G-life project, new networking concepts and solutions are being developed.

For the research of using 6G for medical applications, the communication and the medical side have joined forces: While researchers from the MITI group (Minimally invasive Interdisciplinary Therapeutical Intervention), located at the hospital "Rechts der Isar", focus on the requirements of the medical applications and collecting needed parameters of patients, it is the task of the researchers at LKN to optimize the network in order to satisfy the applications' demands. The goal of this joint research work is to have working testbeds for two medical testbeds located in the hospital to demonstrate the impact and benefits of future 6G networks and concepts for medical applications.

Your task during this work is to implement the communcation network for those testbeds. Based on an existing open-access 5G network implementation, you will implement changes according to the progress of the current research. The results of your work, working 6G medical testbeds, will enable researchers to validate their approaches with real-world measurements and allow to demonstrate future 6G concepts to research, industry and politics.

In this project, you will gain a deep insight into how communication networks, especially the Radio Access Network (RAN), work and how different aspects are implemented. Additionally, you will understand the current limitations and weaknesses as well as concepts for improvement. Also, you will get some insights into medical topics if interested. As in such a broad topic there are many open research questions, you additionally have the possibility to also write your thesis or complete an internship.

• Most important: Motivation and willingness to learn unknown things.
• C/C++ and knowledge about how other programming languages work (Python, etc.) and/or the willingness to work oneself into such languages.
• Preferred: Knowledge about communication networks (exspecially the RAN), 5G concepts, the P4 language, SDN, Linux.
• Initiative to bring in own ideas and solutions.
• Ability to work with various partners (teamwork ability).

Please note: It is not necessary to know about every topic aforementioned, much more it is important to be willing to read oneself in.

In this work, expected result is to enhance the 5G/6G testbed setup with several additional features mainly focusing on the Radio Access Network (RAN). The student is expected to work on the OpenAirInterface (OAI) [1] platform which is the basis of the testbed setup. Expected outcome is to have improvements to the RAN of OAI including but not limited to wireless channel estimation and equalization, Uplink (UL) resource scheduling, and power boosting. More details will be provided after the first meeting.

[1] N.Nikaein, M.K. Marina, S. Manickam, A.Dawson, R. Knopp and C.Bonnet,
“OpenAirInterface: A flexible platform for 5G research,” ACM SIGCOMM Computer
Communication Review, vol. 44, no. 5, 2014.

- Good C/C++ experience

- Medium knowledge on OFDM and Wireless Channel Estimation

- Good Python knowledge is a plus

- Machine Learning understanding is a plus

 Software-Defined Networking (SDN) is a network paradigm where control and data planes are decoupled. The control plane consists on a controller, which manages network functionality and can be deployed in one or multiple servers. The data plane consists on forwarding entities which are instructed by the controller on how to forward traffic.

 A network can be divided in multiple domains in order to ease its management or limit ownership. In multi-domain SDN, each domain has a controller which is responsible for the management. Controllers in different domains cooperate which each other aiming at providing multi-domain end-to-end connectivity.

 In this work, the student will receive an abstract topology representing the multi-domain network. This information has to be used to build a virtual network, that can be used in the testing of different algorithms. The implementation should include a GUI, in order to visualize the topology and interact with the different elements in the network.

Please send your CV and transcript of records.

Basic knowledge on the following:

• Linux
• Networking/SDN
• Python
• Web programming (GUI)

As preliminary results show, Linux TCP/IP Networking Stack introduces a high networking delay. The topic of this work is to perform an empirical study on the Linux socket-based transmission approach and implement a delay measurement workflow based on existing foundations and repositories.

References:

[1] Where has my time gone?

Basic knowledge of 

• Networking and Linux
• C

Future IoT will need wireless links with high data rates, low latency and reliable connectivity despite the limited radio spectrum. Connected lighting is an interesting infrastructure for IoT services because it enables visible light communication (VLC), i.e. a wireless communication using unlicensed light spectrum. This work will aim at developing a tool to perform an economic evaluation of the proposed solution in the particular case of a smart office.

For that purpose, the following tasks will have to be performed:

• Definition of a high-level framework specifying the different modules that will be implemented as well as the required inputs and the expected outputs of the tool.
• Development of a cost evaluation Excel-VBA tool. This tool will allow to evaluate different variations of the selected case study and if possible, to compare different alternative models (e.g., dimensioning) or scenarios (e.g., building types).

- Excel and VBA

Network Function Virtualization (NFV) is becoming a promissing technology in modern networks. A challenging problem is determining the placement of Virtual Network Functions (VNFs). In this work, we plan to implement existing algorithms for embedding VNFs chains in NFV-enabled networks. 

Experience in Python or Java, object oriented programming

The lecture Broadband Communication Networks (Prof. Wolfgang Kellerer) teaches network-related methods of mobile communication: WIFI, 2G to 5G cellular networks, etc. In order to bridge the gap between the methods and real life, innovative teaching concepts shall be developed in form of short podcasts where the students learn in short episodes about wireless communication and networking in day to day scenarios. In the podcasts the student should also be introduced to short exercises they can perform on their own.

In order to support designing these podcasts Pro. Kellerer is looking for a student experienced in programming and maybe in podcasts to help him.

Very good programming skills; experience with podcasts; experience with app programming (on android/ios)

The Software-Defined Networking (SDN) lab offers the opportunity to work with real hardware on interesting and entertaining projects related to the SDN network paradigm. For the next semester, a position is available to assist the teaching assistants for the lab (definition and preparation of the assignments, preparation of the hardware, etc.).

• Solid knowledge in computer networking (TCP/IP, SDN)
• Solid knowledge of networking tools and Linux (iperf, ssh, etc)
• Good programming skills: C/C++, Python