On Sun, Apr 28, 2019 at 09:04:19PM +0200, Martin Weinelt wrote:
We have been using the early noflood and DHCP snooping patches from Linus since roughly 2018/04. They were based on sockmarking packets that should be handled by noflood. This resulted in quite some amount of ebtables rules on our gateways, that marked addresses within DHCP ranges for noflood, since they were very likely already snooped. The result can be seen in graphs I provided to Linus back then, that are now visible at https://www.open-mesh.org/projects/batman-adv/wiki/DAT_DHCP_Snooping#Result.
Though to be fair, I'm expecting similar results with the DAT DHCP snooping + pending DAT cache/DHT split patches. At least the Total->unanswered->ok->last-reply->0...30min. and Total->answered->"via DAT BCAST" in the link should go away then (90.09% of all broadcast flooded ARP Replies in this link).
So while that was the main use-case for me before the discussions we had last year on how to improve DAT, it's now more a minor one for me. Though it would still be nice to have / work towards a zero-broadcast mesh as RFC7772 for instance explains that frequent broadcasts have a quite significant impact on battery power for small, mobile devices. And the noflood option would be a quick and easy option for us for now to achieve this for ARP and other packet types.