Predictability of Chameleon
The goal of this thesis is to investigate the patterns exhibited in the decisions of the Chameleon controller.
The paper Chameleon describes a control path algorithm with the goal of achieving predictable latency and high network utilization. Chameleon utilizes path diversity, priority queueing and recalculating routes to outperform state of the art.
The controller decides if a new flow can be embedded depending on the current network state and the requirements of the new flow. Understanding the decision process of Chameleon is crucial to further improve performance. Therefore, the goal of this thesis is to investigate the decisions of the Chameleon controller. The task of the student is to design a framework to easily generate and store data with Chameleon for further evaluation. After this, the student should evaluate the collected data. The goal here is to find patterns the controller exhibits in its decisions.
Amaury Van Bemten, Nemanja Ðeri?, Amir Varasteh, Stefan Schmid, Carmen Mas-Machuca, Andreas Blenk, and Wolfgang Kellerer. 2020. Chameleon: Predictable Latency and High Utilization with Queue-Aware and Adaptive Source Routing. In The 16th International Conference on emerging Networking EXperiments and Technologies (CoNEXT ’20), December 1–4, 2020, Barcelona, Spain. ACM, New York, NY, USA, 15 pages. https://doi.org/10.1145/3386367.3432879
- Experience with Python
- Experience with Linux command line
- Experience with Java is a plus
Measuring the Throughput of quantized neural networks on P4 devices
Implement a quantized neural network in P4 and evaluate the throughput of feed-forward networks and networks with attention mechanisms on P4 hardware.
Automated Generation of Adversarial Inputs for Data Center Networks
adversarial; datacenter networks
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.
 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.
 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++
Research Internships (Forschungspraxis)
Towards Digital Network Twins: Can we Machine Learn Network Function Behaviors?
Digital Network Twins can help to improve future network operation and management significantly. A Digital Twin of a network is a digital representation that is coupled to the real network. It can be used to perform experiments e.g. to improve the operation of the real network. Running a detailed model of a network as a Digital Network Twin can be quite challenging. Either the computational effort is too high to run the model somehow in real-time or the abstraction level is too high so the Digital Twin does not represent the real network closely enough. Here Machine Learning could be a solution. Instead of accurately modeling the behavior analytically, a machine learning approach observes and learns the behavior of a network and its elements and may lead to a model that is less complex to use for a Digital Twin. Research in this direction is still at its infancy, however.
This research internship should investigate the ability of current machine learning approaches to learn the behavior of network functions. For this task, exemplary network functions, e.g., written in P4, should be set up in a virtual and a hardware testbed. The setup should further contain traffic generators to benchmark network functions and monitoring installations to observe the behavior of the network functions. The collected data should then be used to train machine learning models. The core question is whether how far the behavior of network functions can be learned and abstracted with machine learning models.