Arma 3
Laghisola
Ron4Swat Aug 10, 2020 @ 11:05am
Performance
I have a bit of a strange performance issue with this terrain, which i also have seen on another terrain but do not know whats the reason.

The problem starts when AI is spawned in certain areas, then performance breaks down much more then it is to be expected, being around 1/2 of what it should be.

As far i can tell it has nothing to do with the objects placed in the area, because when i hide them through the "hide terrain objects" feature in eden editor i get the same performance loses.

I experience this for example around Manikaaua, but also in other areas.

I wonder if the cell size of the terrain has something to do with it. It might be better to use only integers as cell size, else there might arrise some issues for whatever reasons, or some optimizations cannot be used.

Then the cell size might be too low for a terrain of this size. For Anizay for example, that is as big as this terrain ( https://steamcommunity.com/sharedfiles/filedetails/?id=1537973181 ) the cell size was eventually raised to 5 because of performance reasons (see changelog).

Don't know how difficult it is to change the cell size of a terrain, but might be interesting to test out different values between releases to see if it makes a difference for the mentioned problem. Testing out if a cell size of 2 or 3 makes a difference would be interesting to see.
< >
Showing 1-12 of 12 comments
Ron4Swat Aug 10, 2020 @ 3:35pm 
Did some more tests and had a interesting finding which does not really resolve the above described performance issue, but showed up another one.

When using the "hiding terrain objects" functionality in eden editor it turns out that deactivating the "Other" objects in a radius of 1000m around Manikaua with same object viewdistance increased the fps around 30-40%. "Other" objects are things like rocks, stones, very small trees/logs.

Hiding Vegatations and walls and fences did not make any difference in fps, even so they were combined around 9000 objects in the 1000m radius, the "Other" objects were around 6500.

This amount of "other" objects seem to be relative alot compared to other terrains. VT7 and RHSPKL for example have around 2500-3500 in a 1000m radius, at least at the places i looked.

Khoramshahr has even more with around 9000 "other" objects in a 1000m radius.



EAGLE  [developer] Aug 11, 2020 @ 5:19am 
If cell size did not resolve the issue, then I have to decrease amount of small stones, logs, bushes.
Ron4Swat Aug 12, 2020 @ 11:31am 
Tested out the first update from yesterday that switched to a 5m cellsize and it had no impact.

With the 2. release later that day i had some positive results. But i also got some runs were the issue was as before, even so the "fix" with and without reduced "other objects" worked reliable through multiple runs before. So it might just have looked as if the "other objects" were the problem, while the real issue is somewhere else.

I still have copies of the older versions of the terrain to make some comparisions if the changes actually did something or not, but it will probably take some time to figure out what is going on.
LuizBarros99  [developer] Aug 16, 2020 @ 1:54pm 
Originally posted by EAGLE:
If cell size did not resolve the issue, then I have to decrease amount of small stones, logs, bushes.
I suspect might mostly be "decoration" items that could be tuned down. (Crates, sacks, barbed wire, ground clutters, medical stretchers, etc.)
EAGLE  [developer] Aug 16, 2020 @ 2:52pm 
Now the terrain has less clutters and middle part of the terrain has less trees & other objects. cell size is 5 (I personally like 2.5 because I can give small and smooth heights to terrain). So if any performance issues remain in the middle part, the reason can be objects such as crates, H-Barriers ...
Last edited by EAGLE; Aug 16, 2020 @ 2:54pm
Ron4Swat Sep 18, 2020 @ 2:48pm 
It's a month since i last posted about this, so might be helpful to give some updates.

First the good news. Neither cellsize or the objects were the problem of the performance issues. So that might help to clear up any remaining questions how much detail to use on your terrain.

The bad news is that i still do not know what exactly causes this but at least figured out how to "fix" the performance issues.

As it turned out after some experimentation was that part of the problems are all the soldiers i spawned in the houses. I eventually found out that disabling their animation changes via disableAI "ANIM" made the performance issues go away.

After some more testing i found out that executing a doStop on all the groups staying in the houses had the same result. I can reliable reproduce that at any time in the mission, executing a doStop on all that groups nearly doubles the server fps and brings it to a normal level which i known from other terrains. Reverting back their behaviour via executing doFollow on the groups cuts the server fps in half again, even so they stay in houses the same way regardless.

I was not able to reproduce this issue in other terrains so far, so there might still be something in the terrain which triggers the issue.
Last edited by Ron4Swat; Sep 18, 2020 @ 2:49pm
EAGLE  [developer] Sep 19, 2020 @ 1:42am 
I'm working on new mask layer, preparing my own dedicated server so I can test and find out how to fix the issue
Ron4Swat Sep 19, 2020 @ 11:59am 
You do not necessarly need a dedicated server for it, its reproduceable on a hosted game too. Its just a more measureable difference on a dedicated server.

On a hosted game the difrference in fps is more around 60% and not the 90-100%,which is still high enough to see the difference.

At this time its probably not useful to invest more of your time into this issue until the issue is confirmed by others. Right now it seems i'm the only one running into this problem. Maybe its not just a combination of AI+the mission but AI+mission+certain hardware or even the OS, who knows.

I will try if i can reproduce the issue on another system too, but as i can already work around this issue as described in previous post, its not that much of a problem anymore.
EAGLE  [developer] Sep 19, 2020 @ 12:07pm 
5.61.28.47

AI spawn in Manikaua, Baltarni, San de Ropa and patrol around/in houses/towers,...

Zeus enabled

Server FPS is ~40-48
In combat mode ~30-48

Server: 4GB RAM, Xeon E5 2620 2.1GHz (may not be exact because it is VPS)
Ron4Swat Sep 19, 2020 @ 12:16pm 
How many AI ? When i test this issue i have around 150-200 simulated in the mission.

Combat mode is not much of importance here as long there are no enemies around and then its natural that fps is lower. Its also important that AI is not moving around to see the issue, because moving AI naturally lead to a lower fps.

Doing a

{doStop (units _x)} foreach allGroups

should increase the fps, executing

{(units _x) doFollow leader _x} foreach allGroups

should decrease it.
Last edited by Ron4Swat; Oct 15, 2020 @ 10:24am
EAGLE  [developer] Oct 15, 2020 @ 9:34am 
Compared with other terrains, not too much differences but yes it does. I can't find what cause this issue, maybe the mask layer (I can't confirm yet).
Ron4Swat Oct 16, 2020 @ 5:08pm 
As you seem to be able to reproduce the issue however not on the same scale i wanted to do some more tests. Suddenly i could not reproduce the issue first.

Moved back to a older version of the mission and suddenly i could reproduce the issue. Strange, something else in the mission must influence if the issue appears.

Turns out that behaviour of the groups is important too. It must be "SAFE" to make this issue appear on full scale. If i have them on "COMBAT" i still see the issue but on a much more subtile scale, around 20% 30% or less.

{_x setBehaviour "SAFE"; (units _x) doFollow leader _x} foreach allGroups triggers a fps decrease of around 50-60% on server side,

{_x setBehaviour "COMBAT"; (units _x) doFollow leader _x} foreach allGroups triggers a fps decrease of around 20-30% on server side,


Just to be sure i want ot mention that these commands have to be executed on server side when using a dedicated server.
Last edited by Ron4Swat; Oct 16, 2020 @ 5:31pm
< >
Showing 1-12 of 12 comments
Per page: 1530 50