[ovs-dev] [flow-compat 7/7] ofproto-dpif: Support differing user/kernel packet parsing support.

Ben Pfaff blp at nicira.com
Wed Nov 16 01:20:00 UTC 2011


On Wed, Nov 09, 2011 at 11:08:11AM -0800, Jesse Gross wrote:
> On Fri, Nov 4, 2011 at 4:43 PM, Ben Pfaff <blp at nicira.com> wrote:
> > Feature #4886.
> > ---
> > ??lib/odp-util.c ?? ?? ?? ?? | ??465 +++++++++++++++++--------------
> > ??lib/odp-util.h ?? ?? ?? ?? | ?? 17 +-
> > ??ofproto/ofproto-dpif.c | ??724 +++++++++++++++++++++++++++++-------------------
> > ??tests/automake.mk ?? ?? ??| ?? ??2 +
> > ??4 files changed, 719 insertions(+), 489 deletions(-)
> 
> I'm still looking through the details of this but I have one high
> level concern that I'd like to bring up first:
> 
> To me, compatibility is not so much about making a feature work even
> if one of the sides doesn't support it but ensuring that existing
> functionality continues to work even if unrelated things are added.
> For example, if one side supports IPv6 but the other does not, then I
> don't necessarily expect IPv6 to work at all unless I upgrade.
> However, if I'm doing L2 switching only and upgrade one half to a
> version that supports IPv6 then I don't expect a change in behavior.
> 
> On the kernel side of things, this expectation is met after this patch.
> 
> If userspace upgrades to support IPv6 but the kernel does not then
> performance will fall off a cliff for IPv6 traffic.  This would likely
> be a pretty big surprise.
> 
> So in that light, it seems to me that the ability to check which
> fields are important is much more than just an optimization and is
> actually a fairly core part of compatibility.

This patch, significantly revised, is now in the "vlan splinters"
series that I just posted:
	http://openvswitch.org/pipermail/dev/2011-November/012960.html
I haven't yet fixed this problem that you bring up, but I will.



More information about the dev mailing list