News for hackers. Stuff that matters.

A comment on Surpassing 10Gb/S over Tailscale.

Re:Surpassing 10Gb/S over Tailscale by jeffbee on Thursday April 13, 2023 @05:17PM (#35562174)
Half-way through the article it just says UDP receive coalescing, once, and never mentions it again. Do they mean interrupt mitigation? If so, using what parameters?
[ Parent | on HN ]
Re:Surpassing 10Gb/S over Tailscale by ignoramous on Thursday April 13, 2023 @06:17PM (#35562884)
My guess is UDP receive coalescing is UDP GRO (generic recv offload) + recvmmsg(2)
[ Parent | on HN ]
Re:Surpassing 10Gb/S over Tailscale by raggi on Thursday April 13, 2023 @07:01PM (#35563337)
Yup, this is referring to GRO.

IIRC we use a contiguous 64kb buffer in the first scatter-gather slot and 128 messages per syscall in the current tuning.

[ Parent | on HN ]
Re:Surpassing 10Gb/S over Tailscale by jordanwhited on Thursday April 13, 2023 @07:13PM (#35563436)
Author here. There was no interrupt tuning performed on the devices under test. UDP receive coalescing was enabled via the UDP_GRO sockopt.
[ Parent | on HN ]