New on LowEndTalk? Please Register and read our Community Rules.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
Comments
Is it LAN or WAN benchmark? I mean, isn't it just bandwidth bottleneck somewhere in between two endpoints?
iperf with 2x cheap vps's below 1€, easily hit 500Mbit.
However my wireguard here is slower, no idea why though.
WAN to WAN. Doesn't seem to be a bandwidth bottleneck as regular iperf saturates the connection easily at 1Gbit.
~800 Mbps
~403 Mbps from Hostsailor AMS Chicken $6/yr
Hitting ~500Mbps on wireguard android to an endpoint in NL, ~120ms away. 550Mbps without VPN to the same endpoint. On a 700Mbps connection.
No tweaks or anything.
All via raw Wireguard or wraps like Tailscale and such?
Jin maek, list your wg servers spcs. You can have 1Gpbs pipe, but if you run on 128MB vps - that would be a problem.
I use a mesh vpn like tailscale, and results vary a lot, depends of you internet speed and protocol you use.
On a lowend connection ~20mbps to download file from my local cloud using https, it is about 700kbps direct to it, and full bandwith with reverse proxy on datacenter (that itself connect to meshvpn to the server).
And on a 1g line, i can only get ~700mbps on sftp, but full 1g on webdav all of that passing through meshvpn.
Just for the record:
Home VPN node:
Intel(R) Core(TM) Ultra 9 185H
32GB RAM
Remote endpoints I tried:
No VPN I easily hit 900+ depending on endpoint.
I haven't tried any meshes, just regular kernel wireguard.
But I dont want to turn this into a support thread as I'm really interested in the perf everyone is getting.
I tried on the wireguard official android app
Multi-hop over WAN:
The GL-AR750 has weak CPU and 100Mbps FastEthernet on both sides, which is believed to be the bottleneck.
Nevertheless, this bandwidth is sufficient for 720p TV streaming.
It really depends on the path between the two points and most of the time there isn’t much you can do. From my PC to one of my proxmox clusters in another country I get around 250mbit/s on 500mbit/s and between the nodes in different DCs but same provider and country it‘s around 450mbit/s on 500mbit/s.
from home (2.5gbps fiber/southern italy) to amsterdam WG installed on cheap VPS (10gbps shared) 800/850 mbps approx. same with netcup root server and hetzner. With tailscale as exit node you need to subtract 20/30% at least on such speed. protonVPN on swiss/NL servers around 1gbps
Host OS? Both linux?
On my LAN I get lower performance going between Windows-Linux machines. But Linux-Linux I can saturate connection and on WAN I can hit maybe 850mbit on a 940mbit peak line.
Home (600 mbps, fiber) to VPS (1.5 gbps shared), Chile.
Client: LXC container, 5900hx 2 core, 512mb ram, debian 12
Server: infofractal.io VPS, Xeon 6244 1 vcore, 1 gb ram, debian 12
iperf3 from vps side (receiver), using UDP
I did some testing using different vps from the same provider and xeon 6244 perform the best. 6262V was ~300 mbps.
EDIT: Also tried same iperf3 without wireguard (pointing to vps public ip) and got 694 Mbits/sec over UDP.
I've got 4.5 gbit/s max with 10G connection and in-kernel WireGuard, single peer single connection.
Make sure you have recent kernel, newer than 6.13 — this version has added GSO support.
Or even better, try the latest 6.19-rc1, there's been numerous speed ups for network in general and UDP in particular.
One of the major slowdown reason which is easily missed are ipatbles/nftables rules. The larger the amount, the slower the connection. You can profile it with
perfand see how much of CPU time is spent inside nf_* functions.This is common with
fail2ban/sshguard: if there are lots of banned IPs, it can slow down the network processing significantly. Both of these add blocking rules in a way which make them execute for each packet. The fix is to check if the connection is already established (conntrack established) and pass it as a first input/forward/output rule.Holy shit, is this the real @ValdikSS -sensei!? Didn't know you were here on LET.
As someone from a heavily censored region I appreciate your research on anti-censorship.
You should come shitpost with us sometime.
This smells like ISP throttling since you get similar speed on 3 different locations, try with openvpn on tcp if you get better speeds.. or mby use @Nyr wireguard script ?