Memory Issues under Wine/Proton 11

Wine/Proton 11 Experimental
Issue is present on previous Wine versions
Linux 7.2.9-arch1-1

The issue is: Client memory grows without stopping over the course of a session. Under extreme circumstances a single client can reach 20gb of ram if allowed to.

From what I am able to tell from my investigation on my own system is that the observed growth was within EVE’s exefile.exe process. Its memory was predominantly private/anonymous dirty memory, while the proton helper process accounted for only a comparatively small amount. The available measurements (linux process interfaces) did not show fsync mappings or GPU mappings accounting for the multi-gigabyte increase. I did not see this behavior in Windows.

It’s observable in game as well. Load in, docked to jita, client spikes to 2.5 then 3gb. Undocking will push this higher and it will continue to grow as long as the session is active. The rate is small for something like sitting on Jita undock and is most noticeable in heavy fleet combat. 20min on the jita undock saw my memory climb from 3.3gb at undock to 4 and it continued to trend upward.

From what I am able to tell, there’s no real difference in the memory consumption rate between DX11 and DX12. Lowest visual settings slows the increase rate and starts from a lower base but does not stop it and the increase is present there too.

I’m curious to see if anyone else has noticed their clients using huge amounts of memory.

This screenshot is 3 minutes of sitting 200km off jita undock just chilling in space.

And closing in on 10min it is still trending upward at a reasonably steady rate.

Sorry, i saw your thread too late. I could try this week to see what my wine reports. Maybe it is a proton issue.

At least i can’t recall to have really experienced any memory problems. I am using wine on debian with DX12. I’ll try to find the time to measure it.

Maybe Jita also is a bad example for testing. It is the most populated system with the most diverse assets to load…so really the worst case and therefore it always has been the slowest and laggiest system.

10 minutes 100km from Amarr VIII (Oris) look like this for me - leveled out below 3.0 GB:

10 minutes 100km from Jita IV-4 - higher as expected, but not racing off like yours.

Also as i wrote…Jita is maybe the worst example with the most assets to be loaded.

Might be a memory leak in proton that is not present in wine.

That’s possible, I’ll try under wine without proton and report back. Thanks for replying!

I’ve kind of kept an eye on this the last week or so. I honestly haven’t noticed any memory creep. Had several multi-hour sessions. Tried proton-experimental, GE 11.7, maybe another.

I have had a few odd clients crash out over the last couple months though and one client is up and the other crashes saying lost connection. It clearly didn’t really lose connection. Those crashes didn’t have any resource monitor up so not sure if it was this.

My PC has 64GB of RAM and 16/32 7950X3D and I had same issue as you.

Similarly I thought it is memory leak, so wanted to test how far it could go, how much it could consume.

Crazy idea came to my mind, increase SWAP from base 64GB to ridiculous value 256GB on my CachyOS (I have 2TB NVME storage, so its not a problem for me). I did some tests and I am able to keep up with 20 accounts opened at the same time, doing mining.

Here is my test scenario: undock from structure, warp to nowhere (like safe spot in system), then warp to busy grid (like moon mining by corp members).
This is how SWAP consumption looked like for me:

start (docked) 34/256
undock 38/256
re-warp 39/256
dock-in 34/256
undock 44/256
re-warp 46/256
dock-in 49/256
undock 52/256
re-warp 53/256
dock-in 54/256
undock 60/256
conduit jump 68/256

after that it stayed more or less at 68/256 GB consumed, it didnt went over 70GB. Game doesnt crash anymore with out-of-memory errors. My RAM consumption during tests was jumping between 45-55-60GB.

If you plan to conduct similar tests, please share your results here.

Thank you for your tests, but why exactly did you think it is a good idea to change the proposed test scenario of the OP where his problem showed? Also you seem to have ignored my tests?

As said above, i think the main problem is Jita where almost any shiptype will fly through in lots of SKINS. That is not happening in your belt, so no constant and heavy asset loading.

So if you repeat that as we two did and stay 100km outside of Jita 4-4 with your characters (preferably in a cloaked ship - for obvious reasons) you will see quite different results.

Maybe you can’t even get your 20 chars to login there or undocked on one machine. Would be interesting to know.

After more investigation, we identified a leak within wine’s low level heap. Patches have been submitted and one was merged to proton-ge latest; wine pr still waiting. This does help but in large fights like the recent amamake throwdown clients were at 15gb.

I suspect there is an eve specific issue with how it interacts with the windows pagefile and some weirdness there while under wine/proton. I have not confirmed this yet, let alone have any idea how to fix it if it’s even possible.

Additionally: enable ksm in the kernel; it will reduce memory consumed by the eve launcher by gigabytes.

Well at least i learned a new feature today that i never heard of before. Will try it out soon.

  1. Check if your kernel supports ksm - which should be true for most distros:

ls /sys/kernel/mm/ksm/

If you get a list of settings, ksm is configured and present in the kernel.

  1. Activate it by issuing:

echo 1 | sudo tee /sys/kernel/mm/ksm/run

You’ll have to configure it as you want depending on your hardware since it does use CPU but the amount is nominal. It is also opt-in by application so be sure you set it for what you want.