On 18 Feb 2025, at 10:58, Roi Dayan wrote:

> On 17/02/2025 10:43, Eelco Chaudron wrote:
>>>> Hi Roi,
>>>>
>>>> Sorry for the late response; I was quite busy with other tasks. However, I 
>>>> did find some time to dive into this piece of code. (I also found some 
>>>> other areas for improvement and will send patches later.)
>>>>
>>>> The cache is nothing more than a neighbor table, and having a port in the 
>>>> output does not provide any benefits. In fact, it adds more confusion, as 
>>>> the port is not related to the entry at all; i.e., it does not represent 
>>>> the egress port. The actual egress port is determined by the OpenFlow 
>>>> rules.
>>>>
>>>> This differs from, for example, the kernel neighbor table, which does show 
>>>> the actual egress port. Given this, I believe it’s best to keep the 
>>>> current output, as it clearly indicates that the packet will exit via the 
>>>> mentioned bridge using the existing OpenFlow rules.
>>>>
>>>> Cheers,
>>>>
>>>> Eelco
>>> Hi,
>>>
>>> Ok we can drop this patch but we thought it could improve output as we 
>>> didn't
>>> really see a scenario were ip is configured on port A while there will be an
>>> openflow rule to force output to port B anyway.
>> Not sure what configuration you are referring to, but for 99%, the port the 
>> system IP is assigned on will not be the output port. Mainly because the 
>> default NORMAL rule will not allow this. In addition, if you are able to 
>> force the output to a physical port with custom OpenFlow rules, this will 
>> only work on DPDK ports with a bifurcated driver.
>
>
> Hi Eelco,
>
> Thanks for the explanation, I was looking it the wrong way.

Don’t worry, it was a good exercise from my side to better understand the code 
;)

//Eelco

_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev

Reply via email to