[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Qemu-discuss] Host high CPU load with network traffic from the guests

From: Malisan, Max
Subject: [Qemu-discuss] Host high CPU load with network traffic from the guests
Date: Fri, 31 Mar 2017 09:28:47 +0000



Lately I have been doing some tests/benchmarks with QEMU and KVM support.

What I noticed is that when I run iperf3, either in server or in client mode, I see a somewhat high CPU load (25-30% on my Intel I5 6300U) on the host from each qemu process (while the guest OS shows almost no load, as I would expect).

I am running all these with a XUbuntu 16.10 (kernel 4.8.0-45) and qemu 2.6.1 with KVM drivers (both were installed using Ubuntu standard packages).


The command I use to start the VM is


                qemu-system-x86_64 \

                                -enable-kvm \

                                -boot menu=on \

                                -smp sockets=1,cpus=4,cores=2 -cpu host \

                                -m 1024 \

                                -vga none -nographic \

                                -drive file="$IMAGE",if=virtio,aio=threads,format=qcow2 \

                                -netdev tap,id=mynet0,ifname=$TAP,script=no \

                                -device virtio-net-pci,netdev=mynet0,mac=$MAC0 \

                                -debugcon file:debug.log -global isa-debugcon.iobase=0x402


Where $TAP is a tap network I had previously created with


                ip tuntap add $TAP mode tap


and bridged to my physical network interface.

My guest VM is being created with Yocto and has a 4.8.x kernel; image format is qcow2 (created by converting the Yocto hddimg with ‘qemu-img convert …’ – for some yet unknown reason the qcow2 Yocto creates doesn’t boot, but I don’t think this is pertinent).

From inside the guest I see (from dmesg) that KVM is being detected as Hypervisor.


My doubt is: am I doing something wrong in the image configuration/startup or the CPU load in the host is somewhat physiological?

Note that the performances with a 1Gbit network test are exactly the same as the host.




reply via email to

[Prev in Thread] Current Thread [Next in Thread]