Feature Description
This document describes how to set up the enablement environment and enable the network multipathing feature on the openEuler OS running on Kunpeng 920 series processors or Kunpeng 950 processors. It also provides guidance on performance testing and troubleshooting.
In Internet service deployments, a single server typically hosts multiple containerized services. When a service process handles network traffic, the CPU core processing NIC interrupts often resides on a different NUMA node than the service process itself, resulting in increased response latency due to cross-NUMA memory access.
The network multipathing feature addresses this by strategically binding NIC queue interrupts to CPUs across different NUMA nodes. By analyzing traffic patterns of specific service processes, it ensures that network traffic of each process is preferentially handled by NIC queues on its local NUMA node, thereby establishing affinity between service processes and their network interrupts.
In heavy-load scenarios, the network multipathing feature processes all network interrupts in a centralized manner. This can lead to CPU overload as a large number of network interrupts are handled by one or a few CPU cores, resulting in performance degradation. In this case, the receive packet steering (RPS) enhancement feature can be enabled to distribute data packets received by NICs to multiple CPU cores for processing. This leverages the multi-core advantage and prevents a single CPU from becoming a bottleneck.
Constraints
NICs must support the Flow Director (FDIR) function. For details, see How Do I Check Whether a NIC Supports the FDIR Function?.
Application Scenarios
The network multipathing feature is applicable to physical machines or containers with high service network loads. It enables NUMA affinity scheduling between service processes and their network interrupts, improving memory access efficiency and service performance by more than 20%.