Tuning Overview

Tuning Process Flow

This section describes how to optimize the NIC performance and use the NIC capability to share the CPU load to improve the performance. In high-concurrency service scenarios, you are advised to use two NICs to reduce the number of cross-chip memory access operations. That is, the two NICs are bound to different CPUs of the server. Each CPU processes only the data of the corresponding NIC. In high-concurrency scenarios, you can also select PCIe x16 cards for NICs.

In a high-concurrency scenario, a large number of requests are sent to the server at the same time or within a short period of time. Each request requires the server to consume resources for processing and providing feedback.

Main Optimization Parameters

Optimization Item	Description	Default Value	When to Take Effect	Kunpeng 916	Kunpeng 920
Adjusting the maximum payload of transaction layer packets (TLPs)	Adjust the maximum value of each data transfer of the PCIe bus.	128B	After the system restarts	Y	Y
Setting the number of NIC queues	Adjust the number of NIC queues.	The number varies according to the OS and NIC.	Immediately	Y	Y
Binding each NIC interrupt to the nearest core	Reduce cross-NUMA memory access.	Irqbalance	Immediately	Y	Y
Aggregating interrupts	Adjust parameters to reduce the number of interrupt processing times.	The number varies according to the OS and NIC.	Immediately	Y	Y
Enabling TCP segmentation offload (TSO)	Send TCP segments to the NIC for processing.	Disabled	Immediately	Y	Y

Parent topic: Tuning the Network Subsystem Performance