Periodic Reporting for period 4 - FlexNets (Quantifying Flexibility in Communication Networks)
Reporting period: 2020-03-01 to 2021-02-28
Communication networks have emerged to become the basic infrastructure for all areas of our society with application areas ranging from social media to industrial production and healthcare. New requirements include the need for dynamic changes of required resources, for example, to react to social events, to shifts of demands or to adoption of new requirements. Existing networks and, in particular, the Internet cannot meet those requirements mainly due to their ossification, and hence limitation in resource allocation, i.e. lack of flexibility to adapt the available resources to changes of demands on a small time-scale and in an efficient way.
In recent years, several concepts have emerged in networking research to provide more flexibility in networks through virtualization and control plane programmability, summarized with the term network softwarization: Software-Defined Networking (SDN), Network Virtualization (NV) and Network Function Virtualization (NFV).
However, a deeper understanding of what flexibility means remains open. In this project, flexibility focuses on the dynamic changes of a network that is characterized by its resources (link rate and node capacities), connectivity (network graph) and its network functions with their related resources, (processing and storage) and their deployment locations. It is the objective of this research to analyze the fundamental design space for flexibility in softwarized networks with respect to cost such as resource usage, performance impact, e.g. latency, and adaptation overhead, e.g. migration. The outcome is the definition of a measure for network flexibility. An analytical model for the definition of such flexibility measure to quantitatively compare different network design choices and to assess the trade-off for flexibility vs. cost has been developed. The design space analysis includes mechanisms for network softwarization as well as general network characteristics such as graph properties and technologies for system optimization based on machine learning techniques. The detailed analysis is based on use cases from different areas including: dynamic resource allocation, function placement, softwarized wireless networks, and resilience.
A fundamental understanding of network flexibility manifested in a quantitative measure and related design guidelines enables all stakeholders in networking to come up with a future-proof system design addressing the dynamic and unforeseeable changes. Thus, it provides a significant benefit to our society as a whole that heavily relies on communication networks. In particular, operators are empowered to react to the emergence of new technologies and regulatory changes. Network flexibility is a key decision factor between network designs, and envisaged as a tie-breaking decisive advantage for a certain network design or technology. For implementation and operation decisions taken by the communications industry, it fills a gap to better understand the options provided by network softwarization and beyond. For research and development, a fundamental understanding of network flexibility supports the analysis of which technical concepts lead to more flexibility in network design to enable generic guidelines for more flexible systems.
In recent years, several concepts have emerged in networking research to provide more flexibility in networks through virtualization and control plane programmability, summarized with the term network softwarization: Software-Defined Networking (SDN), Network Virtualization (NV) and Network Function Virtualization (NFV).
However, a deeper understanding of what flexibility means remains open. In this project, flexibility focuses on the dynamic changes of a network that is characterized by its resources (link rate and node capacities), connectivity (network graph) and its network functions with their related resources, (processing and storage) and their deployment locations. It is the objective of this research to analyze the fundamental design space for flexibility in softwarized networks with respect to cost such as resource usage, performance impact, e.g. latency, and adaptation overhead, e.g. migration. The outcome is the definition of a measure for network flexibility. An analytical model for the definition of such flexibility measure to quantitatively compare different network design choices and to assess the trade-off for flexibility vs. cost has been developed. The design space analysis includes mechanisms for network softwarization as well as general network characteristics such as graph properties and technologies for system optimization based on machine learning techniques. The detailed analysis is based on use cases from different areas including: dynamic resource allocation, function placement, softwarized wireless networks, and resilience.
A fundamental understanding of network flexibility manifested in a quantitative measure and related design guidelines enables all stakeholders in networking to come up with a future-proof system design addressing the dynamic and unforeseeable changes. Thus, it provides a significant benefit to our society as a whole that heavily relies on communication networks. In particular, operators are empowered to react to the emergence of new technologies and regulatory changes. Network flexibility is a key decision factor between network designs, and envisaged as a tie-breaking decisive advantage for a certain network design or technology. For implementation and operation decisions taken by the communications industry, it fills a gap to better understand the options provided by network softwarization and beyond. For research and development, a fundamental understanding of network flexibility supports the analysis of which technical concepts lead to more flexibility in network design to enable generic guidelines for more flexible systems.
The work in this project has focused on the definition of a (mathematically founded) flexibility measure and cost modeling with usage guidelines and detailed use case studies based on experimental systems with a focus on dynamic resource allocation and optimization to evaluate and continuously adapt the flexibility measure and improve network flexibility through emerging concepts such as machine learning and self-driving networks. Results include publications in high class venues, numerous keynote presentations as well as contributions to standardization meetings marking successful initial impact of the results.
For networks, we define that flexibility refers to the ability to timely and cost efficiently support changes in the requirements, by possibly adapting the available network resources, such as flows or topology. Changes arise due to new designers' demands, such as shorter latency budgets or higher availability, or varying traffic characteristics such as sudden shifts of demands. Use case studies performed to validate the proposed flexibility measure include virtual network embedding, network function placement, static and dynamic controller placement, flexible radio access network function spilt, QoS optimization and resilience. Each use case study focuses on a certain system setup that is optimized for certain design parameters. Flexibility is used as an evaluation parameter to compare different design choices such as number of data centers or number of SDN controllers. In order to design flexible network systems, flexibility is used to design and optimize for flexibility. For theoretically sound flexibility analysis, network flexibility as a measure has been defined mathematically together with a comprehensive cost model. The emerging flexibility of networks leads to new opportunities in networking based on a flexible network design. Hence, in the course of the project we have looked into artificial intelligence and machine learning concepts to benefit fully from the provided flexibility. As a result, fundamentally new networking concepts including self-driving networks and network empowerment have been exploited.
For networks, we define that flexibility refers to the ability to timely and cost efficiently support changes in the requirements, by possibly adapting the available network resources, such as flows or topology. Changes arise due to new designers' demands, such as shorter latency budgets or higher availability, or varying traffic characteristics such as sudden shifts of demands. Use case studies performed to validate the proposed flexibility measure include virtual network embedding, network function placement, static and dynamic controller placement, flexible radio access network function spilt, QoS optimization and resilience. Each use case study focuses on a certain system setup that is optimized for certain design parameters. Flexibility is used as an evaluation parameter to compare different design choices such as number of data centers or number of SDN controllers. In order to design flexible network systems, flexibility is used to design and optimize for flexibility. For theoretically sound flexibility analysis, network flexibility as a measure has been defined mathematically together with a comprehensive cost model. The emerging flexibility of networks leads to new opportunities in networking based on a flexible network design. Hence, in the course of the project we have looked into artificial intelligence and machine learning concepts to benefit fully from the provided flexibility. As a result, fundamentally new networking concepts including self-driving networks and network empowerment have been exploited.
In the state of the art, selected aspects of flexibility have been explored for certain network scenarios, a fundamental and comprehensive analysis or even a measure for flexibility is missing. Hence, a measure for flexibility in networks has been proposed and validated through use case studies. In those case studies different network designs have been challenged with “change” requests and the investigated network designs have been optimized to react to the changes in order to evaluate their flexibility with respect to the requests, e.g. changing flow profiles or changing latency requirements. In this way, the flexibility measure is continuously validated and revised.
Significant results include a mathematical definition of flexibility and the related measure. A methodology guides the community of how to apply this fundamentally novel measure. Case study results and benchmark scenarios complement the methodology. Derived from case study analysis, design guidelines indicate how to design network systems for flexibility. In the framework of design guidelines, new technologies and mechanisms for flexibility of networks have been developed and proposed, including the application of machine learning for optimization and dynamic adaptation, network graph related concepts, and the combination of network softwarization concepts for wired and wireless networks.
With respect to the state of the art, the use case studies themselves provide an important contribution due to their proposition and evaluation of potentially new network designs considering flexibility. The new flexibility measure has the potential to create a fundamental new approach in network evaluation, in particular for the new concepts of network softwarization and beyond. During the project standardization bodies such as ETSI NFV and ITU-T have raised their interest in using network flexibility as a measure for standardization.
Significant results include a mathematical definition of flexibility and the related measure. A methodology guides the community of how to apply this fundamentally novel measure. Case study results and benchmark scenarios complement the methodology. Derived from case study analysis, design guidelines indicate how to design network systems for flexibility. In the framework of design guidelines, new technologies and mechanisms for flexibility of networks have been developed and proposed, including the application of machine learning for optimization and dynamic adaptation, network graph related concepts, and the combination of network softwarization concepts for wired and wireless networks.
With respect to the state of the art, the use case studies themselves provide an important contribution due to their proposition and evaluation of potentially new network designs considering flexibility. The new flexibility measure has the potential to create a fundamental new approach in network evaluation, in particular for the new concepts of network softwarization and beyond. During the project standardization bodies such as ETSI NFV and ITU-T have raised their interest in using network flexibility as a measure for standardization.