Method and computer device for affinity binding of interrupts of virtual network interface card

Abstract

The invention discloses a method and a computer device for the affinity binding of the interrupts of a virtual network interface card. The method comprises the following steps of: receiving a request message carried with the affinity policy parameter of the interrupts of the virtual network interface card, and sent by an IaaS resource management system; performing affinity binding on a plurality of virtual central processing units (VCPUs) and a plurality of physical central processing units (PCPUs) in one-to-one correspondence; performing affinity binding on the virtual interrupt of the virtual network interface card and the VCPUs; performing affinity binding on the physical interrupt of the virtual network interface card and the corresponding PCPUs according to the affinity policy parameter.

Description

technical field [0001] The invention relates to the field of communication technology, in particular to a method and computer equipment for a virtual network card to interrupt affinity binding. Background technique [0002] Telecom services are generally computing network-intensive applications, and the software and hardware interrupt processing overhead of the CPU for processing packets sent and received by the network card is very prominent (take the Gigabit network card to send and receive 512-byte long UDP, with a fixed rate of 300Mbps as an example, the hardware and software interrupt processing The overhead can reach 30% of the CPU usage of a single core). [0003] In the physical machine scenario, for a multi-core processor, you can control when and which CPU the NIC interrupt processing runs by setting the affinity between the physical NIC interrupt and the physical CPU. In this way, the telecommunication service program can accurately control the network card inter...

AI Technical Summary

Problems solved by technology

[0006] If there is no definite virtual network card interrupt sending and receiving packet processing scheduling policy, then in the virtual machine environment, it is impossible to predict the impact of virtual network card interrupt sending and receiving packets

like figure 2 As shown, it is assumed that the application program 3 (Application3, APP3) of VM1 is responsible for the process of sending and receiving packets on the network and runs on the virtual VCPU3 (Virtual Central Processing Unit, virtual central processing unit), and VCPU3 is on the PCPU3 (Physical Central Processing Unit , physical central processing unit), if there is no definite policy control, then the processing of the physical interrupt sending and receiving packet of the virtual network card corresponding to APP3 may occur on PCPU2, and the virtual interrupt sending and receiving packet processing of the virtual network card may occur on VCPU1. Thus causing interference to APP2 and APP1 running on VCPU2 and VCPU1 respectively

In addition, if there is no definite policy control of the physical and virtual interrupt sending and receiving packets of the virtual network card, neighbor interference between VMs of the same physical machine may also occur

As shown in the figure below, assuming that APP5 of VM2 is the process responsible for sending and receiving packets on the network and running on VCPU5, and VCPU5 is scheduled to run on PCPU5, then the physical interrupt processing of the corresponding physical network card may be handled by PCPU4 and cause APP4 of VM1 interference

Images