[ovs-dev] [next 29/35] bridge: Move packet processing functionality into ofproto.

Ethan Jackson ethan at nicira.com
Sat May 7 01:03:46 UTC 2011


> I don't yet see how the bond_reconfigure() call is incorrect.
> bond_reconfigure is documented as:
>
>    /* Updates 'bond''s overall configuration to 's'.
>     *
>     * The caller should register each slave on 'bond' by calling
>     * bond_slave_register().  This is optional if none of the slaves'
>     * configuration has changed.  In any case it can't hurt.
>
> so reconfiguring the bond, then adding all of its slaves, seems
> reasonable to me.  That was my intent in the first cut at lib/bond.c,
> as I recall.  Do you see a case where it doesn't work?

Looking at it more closely, I think you are right.  I was worried
about monitoring, but it looks like that case is covered.


> The accounting results are only used for rebalancing decisions, so you
> could just skip it for autopath.

At the moment, there is no reason that autopath actions can't be load
balanced. That said, in the pretty near term, that ability will
probably go away as part of the bonding refactoring that will happen.
I don't think it matters at this point in time.

Ethan



More information about the dev mailing list