Repository navigation
Replies: 1 comment 1 reply
|
Coincidentally we also occasionally see problems with dramatic performance degradation, for us it is when talking to innernet peers that in cloud (Google Cloud for us) - that is also innernet server for us, but also other servers. We haven't been able to fully explain it, but our best explanation is that is indeed something with the MTU. Finding the optimal MTU for wireguard seems to be rather tricky -- it may not be sufficient it is "low enough". See for example https://gist.github.com/nitred/f16850ca48c48c79bf422e90ee5b9d95?permalink_comment_id=4292034#gistcomment-4292034 (the optimial MTU setting is even not symmetrical w.r.t. the 2 link directions). In our case I think Google Cloud doesn't have usual full 1500 bytes on the Ethernet level, it has something like 8 bytes less (for some vlan stuff), which may have an interplay. What kind of protocol above wireguard do you use? If TCP then MSS clamping may be worth exploring. See for example:
It goes without saying that all these are more wireguard than innernet problems, but it's still good to share experiences. Please report if you have any interesting findings! |
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Hi! I have a setup consisting of a central server with a bunch of clients connected.
All the clients are in the same local network, all using gigabit ethernet, connected to a gigabit internet service. The server is running in a datacenter, with a gigabit internet service.
I am seeing low speeds (between 200mbps and 300mbps) between the server and the clients when going through the wireguard interface.
lsmod | rg wireguardshows the module is loaded).iperf3, with the "iperf3 server" on either side, with or without-P5- none of these options change anything to the average speeds reported.iperf3 -M 900or even lower does not seem to improve things, so it looks like the MTU is not the problem.I'm really not sure what is keeping the speed so low. Is the innernet server "special" in term of routing or anything? Could something else bottleneck the speed here?
I'd be glad to run more tests or provide more info to help identify the limitation.
All reactions