Multi-dimensional detection of Linux network congestion based on eBPF

TCP network congestion control algorithm plays a key role in the whole process of TCP data transmission. When network congestion occurs, the quality of network service deteriorates. At present, most servers are operating systems based on Linux. In view of the lack of effective tools for real-time detection and observation of network congestion, this paper proposes a multi-dimensional detection of network congestion on the Linux platform based on eBPF technology, starting with the default congestion control algorithm Cubic in the Linux kernel. The dynamic change of network congestion state machine, the change of current sending window and the exit process of Cubic mixed slow start are detected from the perspective of Linux kernel. Finally, the experimental results show that this method can effectively detect network congestion in Linux, and the real-time multi-dimensional detection of eBPF can detect network congestion earlier and solve the problem of network congestion in the production environment from the perspective of Linux kernel.