High Performance Network Evaluation and Testing

Artur Pilimon

Research output: Book/ReportPh.D. thesis

3188 Downloads (Pure)


This thesis focuses on the testing and performance evaluation aspects of High Speed Networks, covering such areas as Layer 4-7 testing, focusing on protocol-level performance evaluation of long-distance high-speed communication, as well as performance characterization and testing of Data Center Networks (DCNs). From a broad range of open research questions in these areas, a subset of them is investigated in this work, such as: Layer 4 Transmission Control Protocol (TCP) Congestion Control (CC) challenges and algorithmic properties of a high-speed TCP CUBIC extension with additional supporting algorithms, and addressing a scalability aspect of performance evaluation in DCNs from an experimental perspective (defining a methodology for DCN testing at scale). Furthermore, research activities, presented in this work, include testing and analysis of Software Defined Networking (SDN) related performance aspects, namely flow-rule placement in SDN switches and SDN Traffic Engineering (TE) capabilities for DCNs, as well as energy efficiency aspects in DCNs with optical switching. A dominant portion of global data traffic is transported by the TCP protocol. Therefore, a functional and stable TCP connection engine is of paramount importance to ensure reliable and in-order end-to-end data delivery for a diverse range of applications and services. However, the core functionality of the TCP protocol cannot efficiency handle this task alone, and a large set of additional extensions and algorithms are used in combination to enable scalable and adaptive data transmission. In this context, Congestion Control extensions of TCP, such as high-speed CUBIC (used in Linux and Windows 10 Operating Systems), are playing a fundamental role in providing a means of adaptive data transmission in network environments, potentially shared among a large number of different types of flows, as well as preventing such operational network disruptions as congestion collapse. The problem that arises in this situation is that there is still no uniform opinion or agreement on which specific algorithms and extensions enabled with TCP produce the best results in terms of communication stability, adaptability to diverse operational conditions and network environments (e.g., wireless, wired, Internet, DCNs, etc.), resource utilization efficiency, as well as fair bandwidth sharing. As a result, there are many open questions, requiring comprehensive research. Therefore, this work contributes with additional analysis of such aspects as: CC in modern high-speed networks, such as the Internet, as well as algorithmic robustness and packet loss recovery efficiency of high speed TCP CUBIC connections. Data Centers (DCs) have truly become the backbone of the globale conomy, providing a mission-critical infrastructure for a broad range of different applications and services with very diverse Quality of Service (QoS) requirements. Tremendous growth of the global DC IP traffic facilitated active research of new architectural and technological solutions in order to optimize the operational efficiency of the existing infrastructures as well as to build more scalable, energy- and resource-efficient, future-proof DCN architectures and protocols. The question that arises is how do we test and verify that new solutions and innovative research ideas are, indeed, efficient, applicable at large scale and can be transferred to real production DCN environments, while not having access to such facilities for experimental and testing purposes? This is the question that this thesis attempts to answer by proposing a novel hybrid physical-simulated electrical-optical and SDN-controlled DCN testbed for performance and scalability studies and active experimentation. Another important topic, investigated in this thesis, is SDN-related testing and performance evaluation. The growth of scale and complexity of DCN architectures facilitated a need for new optimized ways of resource (compute, storage, networking) management, higher degree of control and programmability of the data plane (network gear) in a more vendor-agnostic, centralized manner. SDN and Cloud Computing paradigms were introduced to address many of these challenges. However, this relatively new technological shift also introduced new challenges in terms of the scalability of the centralized control plane, security concerns, functional capabilities and limitations of the open control interfaces (e.g., OpenFlow), etc. This work focuses on a set of specific SDN-related performance evaluation aspects, such as: performance characterization of a novel flow-rule placement algorithm for SDN switches, focusing on the optimization of resource usage in hybrid (hardware and software) Flowtables, as well as ananalysis of the TE requirements for DCNs and the capabilities provided by SDN in this context. Finally yet importantly, Energy Efficiency aspects in DCNs are explored in this work by evaluating the impact of optical circuit switching, selectively applied on different network layers of a set of considered DCN topologies, such as a traditional three-tier Tree, a Fat-Tree and a Ring-based structure, enhanced by the Wavelength Division Multiplexing (WDM) capabilities. A simulation-based DCN dimensioning is performed and these results are used as an input to a defined power consumption model.
Original languageEnglish
PublisherDTU - Department of Photonics Engineering
Number of pages208
Publication statusPublished - 2018


  • Performance Evaluation
  • High Speed Network Testing
  • TCP Congestion Control
  • Scalability
  • Data Center Network
  • Large-scale testing
  • Hybrid testbed
  • SDN
  • Optical switching

Fingerprint Dive into the research topics of 'High Performance Network Evaluation and Testing'. Together they form a unique fingerprint.

Cite this