> On 3 May 2019, at 20:23, Kevin Darbyshire-Bryant 
> <[email protected]> wrote:
> 
> 
> 
>> On 3 May 2019, at 20:13, Toke Høiland-Jørgensen <[email protected]> wrote:
>> 
>> Kevin Darbyshire-Bryant <[email protected]> writes:
>> 
>>>> On 3 May 2019, at 15:16, Jonathan Morton <[email protected]> wrote:
>>>> 
>>>>> On 3 May, 2019, at 4:55 pm, Kevin Darbyshire-Bryant 
>>>>> <[email protected]> wrote:
>>>>> 
>>>>> Two patches attached - one is a simple variable elimination with no 
>>>>> functional change.  The second changes/simplifies the conntrack tuple 
>>>>> lookup & usage.  I’ve had a play and I don’t think I’ve broken any of the 
>>>>> host fairness BUT it could do with some more testing, that’s where you 
>>>>> come in… 
>>>> 
>>>> Looks like sound logic, as long as it does actually work.  It could be a 
>>>> useful speedup for those small CPE devices which need NAT and 
>>>> host-fairness working.
>>> 
>>> It’s interesting you bring that up - are we sure that ingress host NAT
>>> fairness works in the upstream kernel version of CAKE anyway?  I’m
>>> looking at cake_update_flowkeys(…) and thinking half of it is missing?
>> 
>> No, it's just moved into nf_conntrack_get_tuple_skb(); this was part of
>> the work we did to ensure sch_cake could load without a dependency on
>> the conntrack module...
>> 
>> It does carry over the 'nf_ct_tuple(ct, !hash->tuple.dst.dir);' and the
>> subsequent reversion, though, but I think the logic fits what's in the
>> out-of-tree version?
>> 
>> -Toke
> 
> Ahh! yes I see, thanks. - elixir or my ability to operate elixir was failing 
> earlier.
> 
> Yes and agree the logic follows the out-of-tree…and I can see how my change 
> to it would be applied, assuming it does actually work.
> 

My testing strategy was flawed, the 2nd patch does not work correctly.  Will 
think again.

> 


Cheers,

Kevin D-B

gpg: 012C ACB2 28C6 C53E 9775  9123 B3A2 389B 9DE2 334A

_______________________________________________
Cake mailing list
[email protected]
https://lists.bufferbloat.net/listinfo/cake

Reply via email to