Model-based Analysis of Next-Generation Networks

• Applicants: Professor Dr.-Ing. Georg Carle; Professor Dr.-Ing. Wolfgang Kellerer
• Subject Area: Electronic Semiconductors, Components and Circuits, Integrated Systems, Sensor Technology, Theoretical Electrical Engineering ; Security and Dependability, Operating-, Communication- and Distributed Systems
• Term: from 2018 to 2022
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 397973531

• Final Report Year: 2022

Final Report Abstract

The wide adoption of novel technologies such as Software-Defined Networking (SDN) and Network Function Virtualization (NFV) has shaped how next-generation networks will look. The flexibility of deploying new Network Functions (NFs) on-demand and modifying the network behavior at runtime necessitates the development of a framework that would allow us to predict the performance of such heterogeneous and programmable networks. Therefore, the goal of this project is to provide a basis for the dimensioning and planning of networks and network nodes through the means of the evaluation, combination, and development of models for describing the performance of networks. Throughout this project, we explore a range of different packet processing devices, including programmable switches, SmartNICs, and packet processing software. First, we focus on individual component modeling and leverage the programmability of the network nodes to realize arbitrary packet processing tasks. By combining different processing tasks, we can achieve various network functionalities, ranging from simple packet forwarding to more complex NFs such as routers and loadbalancers. We then implement the NFs in different devices in order to evaluate and model the performance of different components. Additionally, we evaluate the impact of resource virtualization and network hypervisors on performance. Second, we aim to leverage individual component models in order to synthesize network node models. Therefore, we propose a novel methodology to decompose the devices’ functionalities into small building blocks and evaluate their performance in an isolated manner. The network node model is then derived by summing up the individual component models. The correctness of our methodology is confirmed by comparing theoretical values with empirical measurements. Lastly, we combine the obtained component and node models to synthesize a network model. We leverage these models to provide deterministic performance guarantees in environments of low-cost embedded networks and data center networks. The approach taken consists of utilizing a network calculus framework in combination with the component and node models of the devices. We evaluate and compare our models with state-of-the-art solutions to highlight their superiority. Throughout the entire duration of the project, we carry out a cross-sectional work package which focuses on traceability of the executed measurements. For this purpose, we have developed a measurement framework consisting of a structured experimental workflow that allows the creation of reproducible performance experiments in an automated manner. Additionally, we created a framework offering tools to evaluate and visualize the results of the measurements, and to perform an automated model derivation for the device under test.

Publications

• Sdn hypervisors: How much does topology abstraction matter? 14th International Conference on Network and Service Management, 2018
  N. Deric, A. Varasteh, A. Basta, A. Blenk, and W. Kellerer
  
• Cryptographic Hashing in P4 Data Planes. 2nd P4 Workshop in Europe (EUROP4), Cambridge, UK, Sept. 2019
  D. Scholz, A. Oeldemann, F. Geyer, S. Gallenmüller, H. Stubbe, T. Wild, A. Herkersdorf, and G. Carle
  
• Empirical predictability study of sdn switches. 15th ACM/IEEE Symposium on Architectures for Networking and Communications Systems (ANCS’19), Cambridge, UK, Sept. 2019
  A. Van Bemten, N. Deric, A. Varasteh, A. Blenk, S. Schmid, and W. Kellerer
  
• Loko: Predictable latency in small networks. 15th International Conference on emerging Networking EXperiments and Technologies (CoNEXT), 2019
  A. Van Bemten, N. Deric, J. Zerwas, A. Blenk, S. Schmid, and W. Kellerer
  
• On the impact of the network hypervisor on virtual network performance. 2019 IFIP Networking Conference (IFIP Networking), Warsaw, Poland, 2019. IEEE
  A. Blenk, A. Basta, W. Kellerer, and S. Schmid
  
• 5G QoS: Impact of Security Functions on Latency. 2020 IEEE/IFIP Network Operations and Management Symposium (NOMS 2020), Budapest, Hungary, Apr. 2020
  S. Gallenmüller, J. Naab, I. Adam, and G. Carle
  
• Chameleon: Predictable latency and high utilization with queue-aware and adaptive source routing. 16th International Conference on emerging Networking EXperiments and Technologies (CoNEXT), Dec. 2020
  A. Van Bemten, N. Deric, A. Varasteh, S. Schmid, C. Mas Machuca, A. Blenk, and W. Kellerer
  
• A Framework for Reproducible Data Plane Performance Modeling. 16th ACM/IEEE Symposium on Architectures for Networking and Communications Systems (ANCS’21), Dec. 2021
  D. Scholz, H. Harkous, S. Gallenmüller, H. Stubbe, M. Helm, B. Jaeger, N. Deric, E. Goshi, Z. Zhou, W. Kellerer, and G. Carle
  
• Application of Network Calculus Models on Programmable Device Behavior. 33rd International Teletraffic Congress (ITC), Avignon, France, Aug. 2021
  M. Helm, H. Stubbe, D. Scholz, B. Jaeger, S. Gallenmüller, N. Deric, E. Goshi, H. Harkous, Z. Zhou, W. Kellerer, and G. Carle
  
• The pos Framework: A Methodology and Toolchain for Reproducible Network Experiments. The 17th International Conference on emerging Networking EXperiments and Technologies (CoNEXT ’21), Munich, Germany (Virtual Event), Dec. 2021
  S. Gallenmüller, D. Scholz, H. Stubbe, and G. Carle