Hello. I flashed my Flint 2 with the latest vanilla Openwrt, and after researching online I could make it run with Wireguard VPN with PBR and multiple VPN locations.
One thing I tested was the speed, to compare with the Gli.Net firmware, and the VPN speed was much, much slower for the same server and speedteat destination.
Is there a way to get similar speed, or at least more speed, with Wireguard? I enabled HW on the Firewall configs and reduced the MTU on Wireguard to 1280.
Searched online but didn't found anything (only Gen AI suggesting things).
I have 2 routers, so now I went back to the Gli.Net one, but I would love to use OpenWrt for more stability and be outside the vendor firmware (and issues that sometimes appear).
WireGuard itself normally won't use the firewall hardware offload path; that mostly helps routed/NAT traffic, not the crypto work. I would first test one plain tunnel with PBR disabled, then watch top/htop during a speedtest or iperf3 run. If one core is pinned, irqbalance/packet steering can matter more than MTU tweaks. Also try the provider/default MTU again instead of forcing 1280. GL.iNet firmware may have vendor/kernel tweaks too, so matching it exactly on vanilla OpenWrt is not always realistic.
So even with PBR enabled, with irqbalance (I think the remaining was already configured), I'm having much greater speeds (like the same as Flint 2 firmware). Maybe it was temporary, or turning off and now connecting everything made a difference.
So, I'll need to replace the router again with this and check in real life scenarios, if we can say like that. Thank you for the help!
There's no magic here. The throughput promises for WireGuard apply to GL.iNet's stock firmware and only when not using a VPN provider. Even then, it depends on several factors. Another thing, does the ChaCha20-Poly1305 cipher used by WireGuard actually have hardware offload support? From my experience, just set up WireGuard properly, whether it's a VPN provider or a local setup, and you'll get the results.
No there is no hardware offload for ChaCha20-Poly, some CPU's with AVX (X86) or Arm Neon can accelerate it a bit but there is no AES256 like hardware acceleration in a general purpose CPU I know.