> On Jul 3, 2018, at 11:18 AM, Jonathan Morton <[email protected]> wrote:
> 
>> On 3 Jul, 2018, at 1:23 am, Toke Høiland-Jørgensen <[email protected]> wrote:
>> 
>> My hunch is that this has something to do with the way mlx5 uses multiple 
>> receive queues (and thus multiple CPUs). Which is probably different from 
>> veth...
> 
> At this stage I'm pretty confident it has nothing to do with Cake, and 
> everything to do with the Mellanox hardware and driver.  It does strike me 
> that Linux' default handling of multiqueue hardware doesn't map very well to 
> the qdisc interface.

That’s looking highly likely to me too. I’ll stop trying to repro this unless 
we come up with something new to try.

I wonder if enabling the “lockless qdisc” support 
(https://lwn.net/Articles/698135/) we discussed late last year would change 
anything. It should work either way, even with a single qdisc lock, but just 
speculating, at those speeds with multiple cores competing for a single qdisc 
lock I wonder if we’re exposing a problem in the driver or somewhere else.
_______________________________________________
Cake mailing list
[email protected]
https://lists.bufferbloat.net/listinfo/cake

Reply via email to