What Should I Do If the Bandwidth Performance of a Linux ECS Deteriorates?

Symptom

The intranet bandwidth performance of a Linux ECS deteriorates and cannot reach the maximum bandwidth PPS specified in the QoS specifications.

Possible Causes

The ECS runs the Linux 3.10 kernel. Memory overcommitment is not implemented for the RX buffer of the virtio-net driver's receive queue (rq) in the Linux 3.10 kernel. In the packet receiving process, each descriptor's RX buffer must be allocated and released. As a result, the CPU keeps applying for and releasing the RX buffer, and the CPU usage reaches 100%.

During TCP multi-stream bidirectional bandwidth testing, the performance degrades, affecting the multi-stream performance bandwidth. As a result, the bandwidth PPS of the entire ECS decreases.

Solution

Install the index : kernel/git/torvalds/linux.git patch.