Re: [libvirt-users] How can a bridge be optimized?
On Tue, Jun 12, 2012 at 06:14:00PM +0300, Henrik Ahlgren wrote:
The relationship between KVM and QEMU is all too confusing, but
nowadays you should only need to care about the qemu-kvm version (?). 0.12
is quite old and Ubuntu 10.10/maverick no longer receives (security)
updates, so you should consider upgrading if posssible. If upgrading
the whole distro is infeasible, you might be able to install qemu-kvm
and libvirt from a newer Ubuntu release without bringing in too much
dependencies.
Anyone know if I can expect qemu-kvm-1.0.1 to compile and install cleanly
from source? It would be awkward to have to take the system down for a
thorough upgrade, and it's not in a position where security updates are
crucial.
> <interface type='bridge'>
> <mac address='00:16:36:89:65:2e'/>
> <source bridge='br0'/>
> <address type='pci' domain='0x0000' bus='0x00'
slot='0x03' function='0x0'/>
> <model type='virtio'/>
> </interface>
I like to use network type interfaces and "logical" network names, but
this should work fine.
You might want to experiment with different offload settings (ethtool
-k) to see if they make any difference with the igb card. Also, have
you tried measuring the performance between the host and a guest?
From host to guest, a bit above 260 Mb/s. From guest to host, a bit
below 80
Mb. From guest across LAN to another system, 150 Mb +- 10. From across LAN
to guest, 140-180 Mb. From across LAN to host, a consistent 947 Mb.
That's with initial offload settings of:
Offload parameters for eth0:
rx-checksumming: on
tx-checksumming: on
scatter-gather: on
tcp-segmentation-offload: on
udp-fragmentation-offload: off
generic-segmentation-offload: on
generic-receive-offload: on
large-receive-offload: off
ntuple-filters: off
receive-hashing: off
With everything switched off, no change in VM I/O from any perspective.
Whit