On 08/26/15 at 06:19pm, Jiri Benc wrote: > might be a noise. However, there's definitely room for performance > improvement here, the lwtunnel vxlan throughput is at about ~40% of the > non-vxlan throughput. I did not spend too much time on analyzing this, yet, > but it's clear the dst_entry layout is not our biggest concern here.
I'm currently working on reducing the overhead for VXLAN and Gre and effectively Geneve once Pravin's work is in. The main disadvantage of lwt based flow tunneling is the additional fib_lookup() performed for each packet. It seems tempting to cache the tunnel endpoint dst in the lwt state of the overlay route. It will usually point to the same dst for every packet. The cache behaviour if dependant on no fib rules are and the route is a single nexthop route. Did you test with a card that features UDP encapsulation offloads? -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
