Moezilla

There's a scud bombardment of Khasab going on and a lot of aircraft starting up and taxiing. Also @Pillowcat just confirming I'm seeing the same thing with the handles during a run of Scenic-bench.miz (5800X3D/32GB RAM/6750XT/Win11 23H2)

So pathfinding or LOS calculation, maybe? Both of which have constantly been changed and worked on since 2.9.0 due to bug reports from the community, and LOS/detection was specifically called out in the 2.9.6 changelog.

Checked all the logs that were posted for specs: tmz - R7-7800X3D 64GB RAM 4090 Win1123H2 Kahless_uk - R7-5700X 64GB RAM 4070Ti Win1123H2 Rene Coulon - i7-8700K 64GB RAM 2080Ti Win1022H2 A Hamburgler - i7-10700K 32GB RAM 3090 Win1123H2 DragonSoulkin - R5-3600XT 64GB RAM 3060 Win1022H2 Semaphore - i9-14900K 64GB RAM 4090 Win1123H2 Sveppur - R7-5800X3D 64GB RAM 4080 Win1123H2 Bounti30 - i9-9900K 32GB RAM 3090 Win1022H2 SignorMagnifico - i7-14700K 96GB RAM 3080Ti Win1123H2 The_Nephilim - i7-12700K 64GB RAM 4070Ti Win1123H2 mobot - R7-5800X3D 32GB RAM 6750XT Win1123H2 Doesn't seem to be any pattern for AMD vs Intel, RAM amount, GPU vendor, or Windows version. There could be something in the dxdiags, if GPU drivers, BIOS, chipset drivers, etc were also looked at, but that's a bit much. It's interesting that Semaphore with a similar spec to yours, was seeing issues, and had a Core Parking warning in their log. So it could come back to a core scheduling issue but I think this whole thread is in a holding pattern until the next patch.

Use Windows Key+Q to bring up the search bar and type: advanced Click View advanced system settings Click Settings... in the Performance section Click Change... in the Virtual Memory section Set your pagefile up similar to the screenshot, with your pagefile on your fast SSD drive, and no other pagefile on other drives.* Click Set each time after you make changes to a drive. Hit OK and reboot to allow the changes to take effect. * The drive needs to maintain 10% (at minimum) free space after this op.

Crash is due to being out of memory. With 16GB RAM you will probably need to adjust your windows pagefile to 49152 min/max on your fastest SSD to play on busy MP servers running heavy missions on maps like Syria. However, be aware that there will still be a lot of swapping to that SSD. If you can increase your RAM to 32 or 64 without spending a lot, it would be a better solution. Also in options drop your preload radius to 60000 or even 30000.

Using @tmz's miz from the first post I ran back-to-back tests in 2.9.5 and 2.9.7 with the same options.lua. The results were as described before with the 2.9.5 running as expected and spending plenty of time at the set fps cap of 95. The 2.9.7 run struggled to get out of single-digit fps. I then ran the test again with RTSS overlay running in order to use presentmon to gather frame metric data. I think the CPUBusy metric points to what many have already suggested, which is that whatever new "logic" was added in 2.9.6/2.9.7 has resulted in missions that were playable becoming completely CPU-bottlenecked and unplayable. 2.9.5 2.9.7

Yes, it does seem strange to show the Chinook lifting these kinds of assets in promotional material, and then seeing the L118's state in the current build.

Yeah, your mission is a saw-tooth pattern between 6-60 FPS for me in SP, pancake, on 2.9.7.59263 (5800X3D/32GB RAM/6750XT) but when I hit Esc, pausing the simulation, the framerate immediately jumps to my cap of 95 and is flat. I opened the miz in the editor and, apart from the carrier group, there are no moving units that I could see, so it kinda points the finger at the target detection/LOS calculations that saw changes in 2.9.6, rather than route path-finding issues: The log portion for the mission run has several ANTIFREEZE ENABLED lines and one SAME MODEL TIME entry: null 2024-08-28 08:46:27.027 DEBUG LuaGUI (Main): ----onSimulationResume--- 2024-08-28 08:46:27.692 WARNING WORLD (Main): ModelTimeQuantizer: ANTIFREEZE ENABLED 2024-08-28 08:46:28.863 WARNING LOG (2964): 1 duplicate message(s) skipped. 2024-08-28 08:46:28.863 WARNING BACKENDCOMMON (15172): Need to reprocess [DDS] image '/textures/Deck_Normal.dds'. Reason: Conversion requires - expanded pixel size, setting alpha to known value. Source is an 8:8:8 (24bpp) format 2024-08-28 08:46:28.964 ERROR_ONCE DX11BACKEND (15172): texture 'wet_map_map' not found. Asked from '' 2024-08-28 08:46:29.446 WARNING WORLD (Main): ModelTimeQuantizer: ANTIFREEZE ENABLED 2024-08-28 08:46:29.902 ERROR_ONCE DX11BACKEND (7732): texture 'carrier_miniboss_roughmet' not found. Asked from '' 2024-08-28 08:46:30.763 WARNING WORLD (Main): ModelTimeQuantizer: ANTIFREEZE ENABLED 2024-08-28 08:46:34.469 WARNING LOG (2964): 2 duplicate message(s) skipped. 2024-08-28 08:46:34.469 WARNING BACKENDCOMMON (15172): Need to reprocess [DDS] image '/textures/Deck_Normal.dds'. Reason: Conversion requires - expanded pixel size, setting alpha to known value. Source is an 8:8:8 (24bpp) format 2024-08-28 08:46:34.938 WARNING WORLD (Main): ModelTimeQuantizer: ANTIFREEZE ENABLED 2024-08-28 08:46:34.942 WARNING WORLD (Main): ModelTimeQuantizer: SAME MODEL TIME 2024-08-28 08:46:36.488 WARNING WORLD (Main): ModelTimeQuantizer: ANTIFREEZE ENABLED 2024-08-28 08:47:41.096 WARNING LOG (2964): 20 duplicate message(s) skipped. 2024-08-28 08:47:41.096 INFO EDTERRAINGRAPHICS41 (16476): SHM_TERRAIN_SHADOWMAP 2024-08-28 08:47:41.096 INFO EDTERRAINGRAPHICS41 (16476): SHM_TERRAIN_SHADOWMAP 2024-08-28 08:48:10.712 INFO EDTERRAINGRAPHICS41 (16476): surface5 gc() LOD 0 21 squares 2024-08-28 08:48:10.712 INFO EDTERRAINGRAPHICS41 (16476): surface5 gc() LOD 1 38 squares 2024-08-28 08:48:10.712 INFO EDTERRAINGRAPHICS41 (16476): surface5 gc() LOD 2 37 squares 2024-08-28 08:48:10.712 INFO EDTERRAINGRAPHICS41 (16476): surface5 gc() LOD 3 73 squares 2024-08-28 08:48:10.712 INFO EDTERRAINGRAPHICS41 (16476): surface5 gc() 1.180100 ms 2024-08-28 08:48:12.100 INFO EDTERRAINGRAPHICS41 (16476): SHM_TERRAIN_SHADOWMAP 2024-08-28 08:48:12.100 INFO EDTERRAINGRAPHICS41 (16476): SHM_TERRAIN_SHADOWMAP 2024-08-28 08:48:12.311 WARNING WORLD (Main): ModelTimeQuantizer: ANTIFREEZE ENABLED 2024-08-28 08:48:20.181 DEBUG LuaGUI (Main): ----onSimulationPause--- 2024-08-28 08:48:26.854 INFO Dispatcher (Main): Stop 2024-08-28 08:48:27.110 DEBUG LuaGUI (Main): ----onSimulationStop--- false nil nil 2024-08-28 08:48:27.110 DEBUG LuaGUI (Main): ----DCS.isMultiplayer()--- false 2024-08-28 08:48:27.519 INFO TERRAIN (Main): lSystem::CleanScenes()

Crash is from running out of memory. Dxdiag shows several of this type of crash in DCS. One thing to try as a test is to drop visibility range to High, terrain textures to low/min, and your preload to 60000.

Server is single-threaded only, so you will see it load up one core. However, in the latest patches it seems that something such as pathfinding, or unit detection seems to adding more load than before 2.9.6. So missions that ran with medium load in 2.9.5 are now maxing out the core the server is running on.

The mouse polling issue has been here since 2.9.6, iirc. Anything over 500hz can cause the effect you are seeing. Check your mouse software for polling rate (not dpi) and drop it to 500hz or lower and see if it makes the mouse spikes go away.

That one is your decision obvs, but one more data point is that I took your SP_Testing2.trk, renamed it to .miz, and ran it on my machine which does have SRS but does not have Tacview, LotAtc, or pw-dev, and I get smooth fps when I go to F2 view and move the camera around the Hornet.

According to the log-analyzer over on the ED discord, you still have the remnants of a lot of mods in your Saved Games\DCS.openbeta\ folder.

I checked 3 of your logs at random and 2 of their crash signatures match another user's report here, which has been seen by ED.

@BIGNEWY has this one been spotted by the devs/CB testers? The trace looks consistent and there is a similar report from a different user on the ED discord in the dcs-mp-server-admins channel.

https://wiki.hoggitworld.com/view/DCS_enum_world

As of 2.9.6 the S_EVENT_MAX is now 61 while the wiki still shows 46 S_EVENT_SIMULATION_START = 46 S_EVENT_WEAPON_REARM = 47 S_EVENT_WEAPON_DROP = 48 S_EVENT_UNIT_TASK_COMPLETE = 49 S_EVENT_UNIT_TASK_STAGE = 50 S_EVENT_MAC_EXTRA_SCORE = 51 S_EVENT_MISSION_RESTART = 52 S_EVENT_MISSION_WINNER = 53 S_EVENT_RUNWAY_TAKEOFF = 54 S_EVENT_RUNWAY_TOUCH = 55 S_EVENT_MAC_LMS_RESTART = 56 S_EVENT_SIMULATION_FREEZE = 57 S_EVENT_SIMULATION_UNFREEZE = 58 S_EVENT_HUMAN_AIRCRAFT_REPAIR_START = 59 S_EVENT_HUMAN_AIRCRAFT_REPAIR_FINISH = 60 S_EVENT_MAX = 61 Adding these events to the world enum could be useful, I think. Thanks for all your efforts over the years in the DCS community, Grimes.

Left cfrag's miz running for 30mins on a dedicated server. I don't know if the server was in a usable state after the first time the ERROR is reported in the log, but the RegMapStorage system continued to try and recycle group and object IDs every 9-10mins. The memory usage continued to rise all the way through the test, which reproduces OP's bug report that this "memory leak" will eventually crash the server due to lack of resources. Is there a chance there will be a fix for this in the update scheduled for June 26th?

Thanks for the quick reply and the extra detail.

@Pikey does the fix you mentioned here apply for this issue as well?

@Flappie I think you were not around when this one was posted but it would be great if you can try to reproduce the issue with cfrag's miz.

@Flappie I remember you contacted Aerges about this issue when the MiG-29 LODs were added. As of the April 10th update there has been no change yet. Has it been acknowledged and work is ongoing, or does it need another bump?

Reproduced with the Huey track posted by OP

Mag3 did add a new lower detail LOD which is transitioned to at 1000m. They also changed the cull distance from 20000m to 10000m. There could be some edge cases at very high FOV (e.g. 21:9 screens at min zoom), or if the new LOD slider is reduced below 1.0, where the MiG-21 model might be culled while still in visual range. I've seen anecdotal evidence about it but I haven't been playing DCS lately to test it, and the reports about 2.9 MP performance have left me uninterested in trying again until I hear about improvements.

@evilnate @Lasko @TobiasA - do you think it can happen on server startup and first mission load as well, or only when the next mission is loaded to a running server with attached clients? It sounds hard to reproduce, but an important issue to track down and squash.