Bug 473120

Currently, with GSO support enabled, the virtio_net driver allocates maximum sized packet buffers and passes them to the host.

This uses up 20 virtio ring entries, allowing us to supply only 20 packets buffers to the host with a 256 entry ring.

This is a huge overhead when receiving small packets, and is most keenly felt when receiving MTU sized packets from off-host.

To improve the performance obtained with small packets, we have introduced a new buffer allocation scheme upstream whereby the guest allocates small buffers, allowing the ring to hold many more small packets than before.

In the case of large packets, the host can combine multiple buffers together to form a larger logical buffer with no effect on performance.

This is an addition to the virtio_net host<->guest ABI which has recently been added to the upstream net-next-2.6 tree and will be included in 2.6.29:

http://git.kernel.org/?p=linux/kernel/git/davem/net-next-2.6.git;a=commitdiff;h=3f2c31d90327f21d76d296af34aa4ca547932ff4

This scheme allows us to be efficient in our use of ring entries while still supporting large packets. Benchmarking using netperf from an external machine to a guest over a 10Gb/s network shows a 100% improvement from ~1Gb/s to ~2Gb/s. With a local host->guest benchmark with GSO disabled on the host side, throughput was seen to increase from 700Mb/s to 1.7Gb/s.