qemu-discuss
[Top][All Lists]
Advanced

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

Re: [Qemu-discuss] E1000 & virtio will not respect set MTU


From: Mike Lovell
Subject: Re: [Qemu-discuss] E1000 & virtio will not respect set MTU
Date: Wed, 01 Feb 2012 11:22:18 -0700
User-agent: Mozilla/5.0 (X11; U; Linux x86_64; en-US; rv:1.9.2.24) Gecko/20111108 Lightning/1.0b2 Thunderbird/3.1.16

On 01/31/2012 06:25 AM, Markus Kovero wrote:
This happens with Linux 3.1.4, libvirt 0.9.6, qemu-kvm 0.15.1 and Linux 3.2.1, libvirt 0.9.9, qemu-kvm 1.0.

It seems guest OS NIC does not respect MTU that has been set on interface when using e1000 or virtio drivers, everything works well when using Realtek emulation though. Problem can be seen as degraded upload bandwidth due tcp retransmissions because of packets get stripped due mtu limitations. I see this definitely as showstopper bug and it definitely needs some attention.

Guest OS NIC is connected to bridge that is over bond device, have tried to set mtu=1500 on bond devices without any help, default mtu on bond devices is 9000 here, because storage access, but Guest OS bridge is set to 1500 and Guest OS itself is on 1500 MTU link.


hrm. i just did a test with a guest using a virtio and e1000 nic and it respected setting the mtu to 1448. when i set the mtu on the interface to 1448 in the guest, tcpdump never showed a packet larger than 1396. my test used qemu 1.0 compiled from source, linux 2.6.35, and no libvirt. are you using the versions of qemu-kvm that game with the linux distribution you are using? if so, which one?

is there a particular reason why the bridge the guests are connected to is set to 1500 when the upstream link is set to 9000? if the rest of your network supports jumbo frames, why not allow the same for the guests?

mike



reply via email to

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