Messages in this thread | | | From | Jesse Gross <> | Date | Fri, 27 Jan 2012 13:50:49 -0800 | Subject | Re: [patch v4, kernel version 3.2.1] net/ipv4/ip_gre: Ethernet multipoint GRE over IP |
| |
On Thu, Jan 26, 2012 at 9:59 PM, Eric Dumazet <eric.dumazet@gmail.com> wrote: > Le vendredi 27 janvier 2012 à 09:24 +1100, Joseph Glanville a écrit : >> David is correct, the forwarding speed of Open vSwitch is at parity >> with the Linux Bridging module and its tunneling speed is actually >> slightly faster than the in kernel GRE implementation. I have tested >> this across a variety of configurations. > > Thanks for this input ! When was this tested exactly, and do you have > some "perf tool" reports to provide ? > > GRE is lockless since one year or so (modulo how is setup the tunnel as > discovered recently)
The out-of-tree OVS GRE stack (which is separate for practical and historical reasons but I would like to combine with the in-kernel one in the future) uses a few dirty tricks to help performance. I don't know the parameters that Joseph used to test but one that he might be benefiting from here is where we cache the full set of headers to be pushed (GRE/IP/Ethernet) as well any routing/switching decisions. This turns a GRE encapsulation into a simple memcpy if there is enough headroom. It's not something that I would want to propose for more general usage but I think that GSO/GRO for GRE would get at least some of that benefit and would also make things easier when NICs with support for encapsulated offloads start showing up. -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
| |