On Thu, Mar 13, 2025 at 12:21 PM Han Zhou <[email protected]> wrote:
>
>
>
> On Sat, Mar 1, 2025 at 6:38 AM <[email protected]> wrote:
>>
>> From: Numan Siddique <[email protected]>
>>
>> This patch series optimizes ovn-controller to add only relevant
>> datapaths to its "local datapaths" map if the logical topology
>> has a single flat provider logical switch connected to multiple
>> routers with a distributed gateway port.
>>
>> (Patch 3 has a detailed commit message.)
>>
>> v1 -> v2
>> -----
>>   * Rebased to resolve the conflicts.
>>
>> Numan Siddique (3):
>>   controller: Store local binding lports in local_datapath.
>>   northd: Add a flag 'only_dgp_peer_ports' to the SB Datapath.
>>   controller: Optimize adding 'dps' to the local datapaths.
>>
>>  controller/binding.c        | 324 +++++++++++---
>>  controller/binding.h        |   2 +
>>  controller/local_data.c     |  98 ++++-
>>  controller/local_data.h     |  10 +-
>>  controller/lport.c          |  12 +
>>  controller/lport.h          |   4 +
>>  controller/ovn-controller.c |  38 ++
>>  northd/northd.c             |  31 +-
>>  northd/northd.h             |   4 +
>>  tests/multinode.at          | 185 +++++++-
>>  tests/ovn-northd.at         |  59 +++
>>  tests/ovn-performance.at    |   6 +-
>>  tests/ovn.at                | 853 ++++++++++++++++++++++++++++++++++++
>>  13 files changed, 1545 insertions(+), 81 deletions(-)
>>
>> --
>> 2.48.1
>>
>> _______________________________________________
>> dev mailing list
>> [email protected]
>> https://mail.openvswitch.org/mailman/listinfo/ovs-dev
>
>
> Hi Numan,
>
> Sorry for the slow review. I have a general comment for this series. I 
> believe the exact same performance optimization is supposed to be achieved by 
> the below commit:
>
> 22298fd37 ovn-controller: Don't flood fill local datapaths beyond DGP 
> boundary.
>
> The scenarios are mentioned in detail in the commit message of that patch, 
> which is very similar to what you described here.
>
> The only exception was when there are distributed NAT or when the redirection 
> type is set to "bridged". In these cases it will fall back to add peer 
> datapath as local. See SB:Port_Binding:options:always-redirect. Is this 
> something that your patch is targeting? Otherwise I wonder why it didn't work 
> without your patch.

Hi Han,

I wanted to mention in the commit message and in the cover letter that
this patch series is an extension of  the commit 22298fd37. I somehow
missed it.

Yes,  I'm targeting the scenario of distributed NAT,

Numan


> Could you help clarify?
>
> Thanks,
> Han
>
>>
_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev

Reply via email to