Upload a .etl log (from UIforETW) saved right after a timeout, with ‘fast sampling’ and preferably ‘tracing to file’ enabled.
I have done it, but where i find .etl log?
EDIT: Got it but its 550mb
I have started a trace and am currently awaiting a timeout, so I can get the data to you! Thanks so much for responding.
Im sure its not resource problem, i moved it back to other vps and it works fine, but it has lower specs for higher price.
I would like to find where is the problem.
24GB ram
8x EPIC 7281 2,1ghz
300mb/s stable, tried to ping.
All games that i host are having connection timed out problem
Odd, I had the timeout connections only using Onesync, so I think for my case it’s a resource. I also changed all the 32 loops to 64 loops though, so I honestly think it’s a resource, but have no clue where…
Here is my .etl log recorded whilst the timeouts were happening to users on my server.
Oddly, this file seems to have been corrupted - it doesn’t load in WPA.
Are you running on a VPS? It seems the emulated network card used (e1000 instead of a native paravirtualized NIC like virtio or vmxnet) is bottlenecking CPU performance heavily.
I have vps,
AMD EPYC 7281 4x
24Gb ram
ill ask for swap
EDIT: thankss
Um maybe cause my dropbox was somewhat full?
Does this work? https://dropmefiles.com/qYN2E
If not, I may have to share with you what I do when I run the tracer.
Instaled virtio and nothing changed
Just installing wouldn’t be enough - you also have to change the configuration on the VM host or it’ll still use the old PRO/1000 MT.
Vps provider told me that they changed it.
Cant even see e1000 in device manager i can see only virtio, ill send screens when i come home.
k, then run a new trace to see what the main thing is now
Interesting - a lot of time spent in sending UDP packets as well, but this time on Xen virtualization.
It’d be interesting to know why ENet is sending so many separate UDP packets - does neteventlog 1
on clients show a crazy amount of packets being spammed?
(also, a crazy amount of time is spent in querying the system timer - maybe this needs to be done even less often, it seems Xen’s emulation of Hyper-V timer queries might be fairly slow?)
Could i have your discord?
I can also share rdp to make it faster
Or i can jist keep sending logs here
Anyway, @kFxDaKing @OfficialBadger - server build 1574 (or higher) might improve UDP sending by not relying on the kernel anymore for concatenating UDP packets but doing so beforehand.