A Survey of Network Measurement in Software-Defined Networking

Lu Wang, Yan LuProceedings of the 2018 International Conference on Network, Communication, Computer Engineering (NCCE 2018) 2018 · DOI

Abstract

The huge size of network and diversified business characteristics make network traffic difficult to predict and manage. Network measurement is an effective way which can ensure QoS (Quality of Service) and network optimization. However, due to the differences in the implementation rules of equipment (such as switches, routers etc.), it is difficult to achieve accurate and timely network measurement in traditional networks. Recently emerging software defined network (SDN) separates the network control plane from the data forwarding plane and manages the whole network through the centralized controller platform which improves the efficiency of the network management. The new network model adopting unified south bound(SB) interface protocols (OpenFlow is the commonly SB protocols) to access network equipment which shields the different implementation details of the underlying network equipment. The SDN solves the problems that the heterogeneous devices statistical information which is difficult to obtain and greatly simplifies the complexity of network measurement. On the basis of extensive research, this paper introduces the network measurement methods commonly used in SDN including sampling method, link latency measurement and link packet loss measurement method. We believe that the network measurement technology in SDN will greatly optimize the performance and accuracy of network measurement and can provide better guarantee for network quality.

Introduction

In traditional network architectures, visibility is a significant hurdle for operators. Networks are typically composed of heterogeneous hardware—switches and routers from various vendors—each running proprietary protocols. This fragmentation makes it incredibly difficult to obtain a timely, accurate, and granular view of network status. Without a unified data collection method, network managers struggle to ensure Quality of Service (QoS) or optimize traffic effectively. Historically, they have been forced to choose between "active" methods, which inject probe packets that consume bandwidth, and "passive" methods, which require expensive, dedicated monitoring hardware.

Software-Defined Networking (SDN) addresses these challenges by decoupling the control plane from the hardware, providing a centralized platform for network management and measurement.

SDN Architecture and Measurement

The core advantage of SDN lies in its layered architecture, which separates high-level business logic from low-level packet forwarding.

The three-layer architecture of Software-Defined Networking.

As shown in the figure above, the architecture consists of three layers:

  1. Application Layer: Contains business applications that define network behavior.
  2. Control Layer: The centralized SDN controller that manages the network state.
  3. Infrastructure Layer: The physical or virtual switches that forward data.

By using a unified Southbound interface, such as OpenFlow, the controller can interact with heterogeneous devices in a standardized way. This abstraction simplifies the collection of statistical information, which was previously difficult to obtain from diverse hardware.

Measurement Methodologies

SDN measurement research generally focuses on three primary pillars: sampling, latency, and packet loss. These techniques leverage the communication between the controller and switches.

Communication between SDN Controller and OpenFlow Switch.

Sampling Methods

Standard periodical sampling involves polling switches at fixed intervals. While simple, it can miss traffic bursts or create unnecessary overhead during stable periods. Adaptive sampling offers an "intelligent" alternative by adjusting polling frequency based on flow rates or throughput changes. This ensures high-resolution data during volatility while maintaining low overhead during quiet periods.

SDN allows for clever latency estimation using standard OpenFlow messages like PACKET_OUT and PACKET_IN. The controller sends a packet to a source switch, which forwards it to a destination switch, which then returns it to the controller. By calculating the total round-trip time and subtracting known controller-to-switch delays, the system can isolate the exact latency of the link between switches. This often requires a calibration value to account for the time the controller takes to process messages.

Packet loss can be measured through flow statistics polling or graph theory approaches. Tools like OpenNetMon compare the number of bytes sent by a source switch versus those received by a target switch. For complex topologies, Euler circuits can be used to measure two-way link loss, providing a holistic view of network health.

Key Insights and Results

The transition to SDN simplifies the measurement stack by moving from vendor-specific CLI scraping to standardized OpenFlow queries. Key findings in the field include:

  • Efficiency through Intelligence: Adaptive polling significantly balances the trade-off between measurement accuracy and system load.
  • Low-Intrusion Monitoring: Methods like the Efficient Passive Lightweight Estimator (EPLE) demonstrate that high-accuracy packet loss detection is possible without the heavy overhead of traditional probe-based systems.
  • Unified Interfaces: Standardized Southbound protocols shield the management layer from implementation details, allowing for "write-once, monitor-anywhere" tools.

Limitations and Future Work

Despite its advantages, SDN measurement faces challenges. Centralizing measurement tasks can place a heavy burden on the SDN controller, potentially creating a bottleneck that degrades network performance. Furthermore, as networks scale to "huge sizes," even efficient adaptive sampling methods face limits. Future research is expected to focus on optimizing non-intrusive monitoring techniques and integrating measurement data more deeply into automated network optimization loops.

Summary updated

Abstract

The huge size of network and diversified business characteristics make network traffic difficult to predict and manage. Network measurement is an effective way which can ensure QoS (Quality of Service) and network optimization. However, due to the differences in the implementation rules of equipment (such as switches, routers etc.), it is difficult to achieve accurate and timely network measurement in traditional networks. Recently emerging software defined network (SDN) separates the network control plane from the data forwarding plane and manages the whole network through the centralized controller platform which improves the efficiency of the network management. The new network model adopting unified south bound(SB) interface protocols (OpenFlow is the commonly SB protocols) to access network equipment which shields the different implementation details of the underlying network equipment. The SDN solves the problems that the heterogeneous devices statistical information which is difficult to obtain and greatly simplifies the complexity of network measurement. On the basis of extensive research, this paper introduces the network measurement methods commonly used in Sandwich including sampling method, link latency measurement and link packet loss measurement method. We believe that the network measurement technology in SDN will greatly optimize the performance and accuracy of network measurement and can provide better guarantee for network quality.
BibTeX
@inproceedings{wang2018survey,
  author = {Lu Wang and Yan Lu},
  title = {{A Survey of Network Measurement in Software-Defined Networking}},
  booktitle = {Proceedings of the 2018 International Conference on Network, Communication, Computer Engineering (NCCE 2018)},
  year = {2018},
  doi = {10.2991/ncce-18.2018.17},
}
Figure