On Mon, Dec 09, 2013 at 03:55:00PM +0200, Michael S. Tsirkin wrote: > On Mon, Dec 09, 2013 at 01:42:30PM +0100, Stefan Hajnoczi wrote: > > So the effect of batching should be relatively small on latency. In > > fact, it's almost like sendmmsg(2)/recvmmsg(2) but using a > > one-packet-at-a-time interface. > > > > Does this sound right? > > > > Stefan > > Why would it be small? Consider a queue of 256 packets. > You are sending out a single short packet, followed > by a burst of 255 larger packets. > the single packet is not transmitted until qemu completes > processing 255 larger ones.
Seems like my intuition is wrong. I figured QEMU processing those 255 packets is quick. If zero-copy is possible then we just need to put the address/length into a ring before we flush the packets to the host kernel. The netperf results do show a regression so we need to understand that better. Stefan