From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:51176) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1f3y5f-0006rI-Vr for qemu-devel@nongnu.org; Thu, 05 Apr 2018 02:09:44 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1f3y5e-00004r-TY for qemu-devel@nongnu.org; Thu, 05 Apr 2018 02:09:43 -0400 MIME-Version: 1.0 From: Prakash B Date: Thu, 5 Apr 2018 11:39:41 +0530 Message-ID: Content-Type: text/plain; charset="UTF-8" Subject: [Qemu-devel] Virito devices bridge vs Macvtap performance List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: qemu-devel@nongnu.org, qemu-arm@nongnu.org, Peter Maydell , Eric Auger , Sunil Goutham , bprakash@caviumnetworks.com, berrange@redhat.com Hi, I am doing kvm guest network performance analysis to compare relevant alternatives for virtio devices on Arrach64. I have generated bridge tap and macvtap performance numbers on guest 10G interface connected back to another 10G on other host. The results for guest with "macvtap" vs "bridge" are marginally differed and the difference is +/- 3%. As per my understanding on macvtap, expected better performance values with macvtap. -------------------- -------------------- for macvtap device:- -netdev tap,vhost=on,vhostforce=on,id=macvtap_netdev,fd=3 3<>/dev/tap$(cat /sys/class/net/macvtap1/ifindex) \ -device virtionet-pci,netdev=macvtap_netdev,id=net0,mac=$(cat /sys/class/net/macvtap1/address) for Bridge tap device:- -netdev type=tap,vhost=on,vhostforce=on,id=tap0,ifname=tap0,script=no,downscript=no -device virtio-net-pci,netdev=tap0,mac=12:03:04:05:06:07 -----------------------< End of commands> ------------------------ Anyone please tell me if I missed anything. Other information. ------------------------- Kernel sources : 4.14.28 qemu-aarch64 : 2.11.91 Did the same experiments with ThunderX vnic and Intel 10G nic and results are as mentioned above. Tried macvtap in bridge and passthru modes. Tell me if you need further investigations or more data . thanks, Prakash B