[ovs-dev] [PATCH 01/16] tunneling: Add support for tunnel ID.
blp at nicira.com
Wed Apr 14 21:16:46 UTC 2010
On Wed, Apr 14, 2010 at 05:10:08PM -0400, Jesse Gross wrote:
> On Wed, Apr 14, 2010 at 3:34 PM, Ben Pfaff <blp at nicira.com> wrote:
> > I just know that changing the argument order here is going to confuse me
> > later. Do you really want to do that? I kind of consider 'rule' here
> > to be like a 'this' argument in C++, so that it should go first.
> The reason why I changed it is because it seems inconsistent with other
> functions where inputs are first and outputs are last. I guess if you
> continue the C++ analogy, this function is a constructor and it is returning
> the object, not taking it as a 'this' argument.
OK, not worth arguing about.
> More generally, as long as we have a static ordering we're always going to
> have poor performance in some situations where it could be better given a
> different ordering. My impression is that given a large number of flows the
> vast majority are going to be of the same form (have the same wildcards).
> This means that if we tracked the types of flows we could self-optimizing
> at runtime and hit the hash table in the majority of what I would expect are
> real world situations.
Yes, I think you are right, that is what we will have to do. I also
have some notions about completely different classifier structures but I
don't when I will have time to experiment with them.
More information about the dev