MODELS: performance MODELing of Software-defined data center networks
Security and Dependability, Operating-, Communication- and Distributed Systems
Final Report Abstract
We proposed the Descartes Network Infrastructure (DNI) approach to flexible run-time performance prediction in data center networks. The DNI modeling formalism captures system behavior at run time and offers different modeling granularities to represent a system at the required level. By offering flexible medium-detailed modeling granularity, the modeling formalism abstracts low level details and focuses on the relevant performanceinfluencing factors. The DNI meta-model has a generic character and is not bound to a given network technology or a protocol. The meta-model offers flexibility to specify the system at different levels of abstraction according to the available data and user needs. The DNI meta-model provides generic modeling entities to support the majority of existing and future network technologies including Software-Defined Networking (SDN) and Network Function Virtualization (NFV). The formalism provides generic support for NFV and medium-detailed support for SDN. The medium-detailed support for SDN supports novel prediction scenarios (e.g., using software-based forwarding tables) and abstracting the factors that negatively influence the overall system performance. The proposed DNI network deployment model allows one to connect the DNI and DML meta-models. It supports mapping the network models into architecture-level descriptive software models so that a complete data center landscape can be modeled. The integration with DML allows capturing behaviors from the computing and software domains, which are not explicitly modeled in the DNI model. Examples of such behaviors include software bottlenecks, server virtualization, and middleware overheads. The integrated descriptive models offer a richer view of the system and are prepared for future integration of transformations and solvers. DNI models have a descriptive nature, which means that they store information about the network infrastructure without providing any means to predict the network performance under different conditions. We provided a flexible way to transform a DNI model into multiple predictive models using model-to-model transformations to enable performance prediction. Each model transformation (or a chain of multiple transformations) contributed in this project enables solving a DNI model by generating a predictive model. The predictive models vary in size and complexity, depending on the amount of data abstracted in the model and the transformation process. To build a DNI model manually, a modeler needs to acquire data about the network. This includes the topology, configuration, and traffic. Some of this information may be cumbersome to obtain and model in DNI manually. For example, the traffic profiles are challenging to extract manually due to the high amount of data transmitted over the network in a short period. For this, we provide methods to support the user in the modeling process by semi-automated extraction of the network traffic part of a DNI model. The DNI approach is the first approach that flexibly bridges the gap between the high-level black-box performance models and the fine-grained, low-level network simulations. We filled this gap by proving different graybox models that capture the internal system structure. This allows investigating various system reconfigurations at different granularities while simultaneously benefiting from the variety of solving times without interrupting the operation of the modeled system. The DNI approach provides benefits to network users and administrators by enabling the analysis of the impact of changes in: the network structure, the network configuration, and the workload profile/intensity. The approach proposed in this project provides a solid basis for the analysis and management of data center network resources in an automated resource management process that may reconfigure the system dynamically to adapt to the future expected conditions. Summarizing, the project makes valuable contributions to support a flexible data center network performance prediction at run time. Although the focus of this project is on SDN-based networks within data centers, the approach can also be applied to classical, non-SDN-based networks in data centers. As part of the project, we have also investigated networks in the context of the Internet of Things (IoT) applications. Especially due to the high scaling of the number of communicating devices in IoT applications, performance prediction is of particular importance here.
Publications
- Automated Extraction of Network Traffic Models Suitable for Performance Simulation. In Proceedings of the 7th ACM/SPEC International Conference on Performance Engineering (ICPE 2016), pages 27–35. ACM, 2016
Piotr Rygielski, Viliam Simko, Felix Sittner, Doris Aschenbrenner, Samuel Kounev, and Klaus Schilling
(See online at https://doi.org/10.1145/2851553.2851570) - Enabling Fluid Analysis for Queueing Petri Nets via Model Transformation. Electronic Notes in Theoretical Computer Science, 327:71–91
Christoph Müller, Piotr Rygielski, Simon Spinner, and Samuel Kounev
(See online at https://doi.org/10.1016/j.entcs.2016.09.024) - Modeling and Prediction of Software-Defined Networks Performance using Queueing Petri Nets. In Proceedings of the Ninth International Conference on Simulation Tools and Techniques (SIMUTools 2016), pages 66–75, August 2016
Piotr Rygielski, Marian Seliuchenko, and Samuel Kounev
- Performance Analysis of SDN Switches with Hardware and Software Flow Tables. In Proceedings of the 10th EAI International Conference on Performance Evaluation Methodologies and Tools (ValueTools 2016), October 2016
Piotr Rygielski, Marian Seliuchenko, Samuel Kounev, and Mykhailo Klymash
- Flexible Modeling of Data Center Networks for Capacity Management. PhD thesis, University of Würzburg, Germany, 2017
Piotr Rygielski
- Performance assessment of cloud migrations from network and application point of view. In Proceedings of 9th EAI International Conference on Mobile Networks and Management (MONAMI 2018), December 2017
Lukas Iffländer, Florian Wamser, Christopher Metter, Phuoc Tran-Gia, and Samuel Kounev
(See online at https://doi.org/10.1007/978-3-319-90775-8_21) - The Vision of Self-aware Reordering of Security Network Function Chains. In Proceedings of the 2018 ACM/SPEC International Conference on Performance Engineering, ICPE ’18, page 1–4, New York, NY, USA, 2018. ACM
Lukas Iffländer, Jürgen Walter, Simon Eismann, and Samuel Kounev
(See online at https://doi.org/10.1145/3185768.3186309) - Modeling of Aggregated IoT Traffic and Its Application to an IoT Cloud. Proceedings of the IEEE, 107(4):679–694, April 2019
Florian Metzger, Tobias Hoßfeld, André Bauer, Samuel Kounev, and Poul. E. Heegaard
(See online at https://doi.org/10.1109/JPROC.2019.2901578) - Performance oriented dynamic bypassing for intrusion detection systems. In Proceedings of the 2019 ACM/SPEC International Conference on Performance Engineering, ICPE ’19, page 159–166, New York, NY, USA, 2019. ACM
Lukas Iffländer, Jonathan Stoll, Nishant Rawtani, Veronika Lesch, Klaus-Dieter Lange, and Samuel Kounev
(See online at https://doi.org/10.1145/3297663.3310313) - A Simulation-based Optimization Framework for Online Adaptation of Networks. In Proceedings of the 12th EAI International Conference on Simulation Tools and Techniques (SIMUtools), SIMUtools 2020, August 2020
Stefan Herrnleben, Johannes Grohmann, Pitor Rygielski, Veronika Lesch, Christian Krupitzer, and Samuel Kounev
(See online at https://doi.org/10.1007/978-3-030-72792-5_41) - Model-based Performance Predictions for SDN-based Networks: A Case Study. In Proceedings of the 20th International GI/ITG Conference on Measurement, Modelling and Evaluation of Computing Systems, MMB 2020, March 2020
Stefan Herrnleben, Pitor Rygielski, Johannes Grohmann, Simon Eismann, Tobias Hossfeld, and Samuel Kounev
(See online at https://doi.org/10.1007/978-3-030-43024-5_6)
