The performance of a network can be measured through several key factors: capabilities, availability, throughput, and security.
Capability represents the ability to interface with a communication network for the purpose of communicating data to or from an IoT device, while the term ‘availability’ means the property that data or information is accessible and usable over time.
Throughput is a measure of how many units of information a system can process in a given amount of time across physical or logical links. Lastly, security addresses the ability of the network to ensure identification, data confidentiality, and the integrity of information in transit through the network.
Evaluating the overall performance of a digital network requires aggregating these metrics across all the transmission channels involved.
Starting from this view, our work aimed to compare the throughput of a traditional underlay network with that of an overlay DaaS network.
To achieve this goal, we developed a loopback testing technique between two nodes, a source and a destination, and assessed the differences in transmission performance over heterogeneous communication channels.
However, we quickly realized that conventional benchmarking tools, used for IP networks, are inadequate for measuring the performance of overlay architectures. Tools such iperf, netperf, and nuttcp are designed around IP standards (IETF RFC 791) and rely on subprotocols like (IETF RFCs 792/777), which are optimized within the network stack itself. As a result, these tools bypass certain layers of the OSI model (ISO 7498), meaning that the measured packets do not traverse the entire protocol stack up to the application layer.
For overlay networks, which often operate at higher layers and rely on heterogeneous lower-layer implementations for transport protocols, this results in incomplete or non-comparable measurements.
To address these limitations, we developed a custom benchmarking tool called DSPerf (https://github.com/sebyone/dsperf).
This tool influence the native communication APIs of IPv4 protocol, along with the APIs provided by LibDaaS (https://daasiot.sebyone.it/download-2/) and the INET4 DaaS driver, which are compiled to operate over the same IPv4 protocol.
By directly interacting with the application and transport layers, DSPerf provides accurate and comparable throughput measurements for both underlay and overlay scenarios.
DSPerf is able of measuring a wide range of parameters, such as total data block size, packet length, header to payload efficiency, packet loss, and transmission rate expressed both in Mbps (byte per second) and pps (packets per second). This level of detail make its possible to analyze the efficiency of network communications and identify potential bottlenecks.
To validate the reliability of DSPerf, we conducted a comparative analysis against results obtained using iperf, netperf, and nuttcp. These tools, despite being the industry standard, provide only partial insights for overlay networks. Our tests were performed on a simple IEEE 802.11b network composed of embedded commercial hardware, chosen for its uniform characteristics supporting multiple networking technologies.
To improve statistical accuracy, each experiment was repeated 400 times with varying data block sizes,
The analysis of the collected data confirmed that DSPer offers consistency of results with traditional tools, with an additional advantage: its measurements are free from the periodic peaks observed in iperf, which are caused by characteristics processing intervals. This difference stems from the DSPerf continuously calculates total transfer time, rather than estimating it at predefined intervals.
When comparing throughput and total transfer times, we observed that DSPerf aligns with the accuracy and repeatability of existing benchmarks while providing a more complete view of the overlay stack.
In conclusion, our experiments demonstrate that DSPerf is a reliable and accurate tool for evaluating network performance across both underlay and overlay layers. By addressing the limitations of existing solutions, DSPerf proves particularly useful for modern network architectures, where hybrid and application-layer protocols are becoming increasingly prevalent. As overlay technologies continue to evolve, tools like DSPerf will play a crucial role in enabling precise, reproducible performance measurements and in helping engineers optimize complex communication infrastructures.
Lascia un commento