A Coloring-Based Packet Loss Rate Measurement Scheme on Network Nodes
Shuhe Wang,Rui Han,Xu Wang
DOI: https://doi.org/10.3390/electronics13234692
IF: 2.9
2024-11-28
Electronics
Abstract:Network measurement is an efficient way to understand network behavior. Traditional measurement techniques focus on internet protocol (IP) networks, where the processing capacity of network nodes is limited and primarily dedicated to packet forwarding. As a result, these techniques typically rely on end hosts or external systems to analyze traffic and evaluate network performance. This reliance introduces several challenges, such as increased measurement latency and scalability limitations, particularly in large-scale networks. With the emergence of next-generation internet architectures, especially information-centric networking (ICN), network nodes have gained enhanced capabilities, enabling measurement tasks to be performed directly at these nodes. This paper proposes a distributed measurement scheme where network nodes collaborate to monitor the packet loss rate on the intermediate link. By setting an unused bit in the packet header, the upstream node "colors" the packets into different color blocks. The minimum duration of each block is determined by the degree of reordering on the link, and the number of packets in each block must be a power of two. The downstream node recognizes blocks, assigns packets to the right block, and deduces the original number of packets for each block to calculate packet loss. Moreover, the upstream node adjusts the number of packets in each block based on the packet transmission rate on the link, aiming to balance measurement accuracy and frequency. A P4-based implementation on a BMv2 software switch is presented to demonstrate the feasibility of the proposed scheme. Simulations show that this scheme improves measurement accuracy and is more robust against packet reordering. Additionally, the proposed scheme maintains relatively low network overhead and, at higher measurement frequencies, exhibits the lowest overhead compared to existing methods.
engineering, electrical & electronic,computer science, information systems,physics, applied