[ovs-dev] [PATCH 2/8] vswitchd: Limit number of ports to 65335.

Ethan Jackson ethan at nicira.com
Mon Jan 31 18:42:47 UTC 2011


After some off-list discussion with Ben and Jesse's comments below I
think the most prudent thing to do is to drop this patch altogether,
and remove the assumption of 16 bit port identifiers from the LACP
patch.  Unless anyone objects, that's what I'm going to do.

Ethan


On Sun, Jan 30, 2011 at 8:25 PM, Jesse Gross <jesse at nicira.com> wrote:
> On Sun, Jan 30, 2011 at 8:14 PM, Ethan Jackson <ethan at nicira.com> wrote:
>> I agree that we should eventually move to 32bit port numbers.
>> However, I think until we make the move consistently throughout the
>> code base it's better to have bridge.c reflect what is actually
>> supported.  Most notably I'm worried about how the bridge interacts
>> with mac_learning.  I haven't done a deep inspection of that code, but
>> mac_learning_learn takes uint16_t port numbers.
>
> It's clear that we don't support 32-bit port numbers now as a number
> of places assume 16-bit, so it's not bad to put a limit on it for the
> time being.
>
>>
>> LACP only allocates 16bits for each "aggregator key" (one per port in
>> ovs case).  So we can have at most 2^16 LACP enabled bonds.  I use
>> port_idx as the identifier in the future patches, but I'm fine with
>> dropping this patch and generating our own uint16_t id number for LACP
>> ports if you think that's preferable.
>
> It would be nice to avoid adding new assumptions on ports being 16-bit.
>




More information about the dev mailing list