On Wed, Nov 30, 2022 at 9:20 PM Thilak Raj Surendra Babu
<[email protected]> wrote:
>
> For my understanding, is part being woken up by a RX interrupt not desirable?

Latency is one thing that could be enhanced with Rx interrupt.
Maybe you have another usecase that would make Rx interrupts a better
fit than simply sleeping.
If so, please share details.


On our side, we are focusing on power savings atm as big deployments
put a high priority on this topic.
So Kevin approach seems more interesting, at least as a first step,
because we get some power saving even at low rates.


> I am under the assumption that all NICs do support rx-interrupts through the 
> fd.

In theory, I would assume too that we can have rx interrupts for hw
drivers (unless we use some old hw, or because of some hw bug...).
If hw is accessed through vfio, we get eventfds mapped to MSIX.
Drivers like mlx (implemented on top of rdma api) provide some fd too.

Even so, you'll see that a fair number of DPDK drivers do not have support.
As for the reason, it may be by lack of interest so far.


> Will take a look at Kevin's patch and try it out as well.

Thanks.


-- 
David Marchand

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

Reply via email to