Deze pagina is nog niet beschikbaar in uw eigen taal. We werken er hard aan om meer taalversies toe te voegen. Bedankt voor uw steun.

On this page

Show all

Help Center/ Bare Metal Server/ FAQs/ Network and Security FAQ/ What Do I Do If the Network Performance Is Poor When 128 Ethernet Network Devices Are Bound to a BMS?

What Do I Do If the Network Performance Is Poor When 128 Ethernet Network Devices Are Bound to a BMS?

Updated on 2023-07-10 GMT+08:00

Symptom

The network performance is poor when multiple Ethernet network devices (for example, 128 single-queue NICs) are bound to an EulerOS 2.3/Euler OS 2.5 BMS. This is because the interrupts of all the 128 NIC queues are processed by the same CPU, causing a critical bottleneck of the CPU and affecting the network performance. The root cause is that --hintpolicy=subset is set for irqbalance of the Euler OS 2.3/Euler OS 2.5 BMS. To solve the problem, change it to --hintpolicy=ignore as instructed in Procedure.

NOTE:

hintpolicy is a policy used by irqbalance to balance interrupts across CPUs based on affinity_hint of each interrupt. affinity_hint indicates the CPU affinity of an interrupt. The value of hintpolicy can be:

  • exact: irqbalance never violates affinity_hint.
  • subset: irqbalance distributes interrupts to a subset of affinity_hint.
  • ignore: irqbalance completely ignores affinity_hint.

Procedure

The following operations use EulerOS 2.3 as an example:

  1. Log in to the BMS as user root.
  2. Run the following command to open the /etc/sysconfig/irqbalance file:

    vi /etc/sysconfig/irqbalance

  3. The original setting is --hintpolicy=subset.

    Change the setting to --hintpolicy-ignore.

Feedback

Feedback

Feedback

0/500

Selected Content

Submit selected content with the feedback