Perfomance drop + log spam in latest 16-02-22 dev patch

[Jan-H 2]

Anyone else noticing:

(a) ARMA3.log spamming with "...bla sees blub..." entries

(b) 10-20% loss of performance in large, vehicle rich scenarios even with -nologs (due to the first)?

 

Since I have been working for more than 4 years on single scenario on the Fallujah map (sort of Blackhawk down, but with AI smartly ordering CGs and adjusting to the situation), I have a good feeling for the FPS (and its scatter range) in that scenario. I have been fighting with pathfinding of AI in the "hotel" type buildings, and spent uncounted hours in trying to get AI groups of vehicles to propperly navigate the twin roads and lately been converting scheduled grp-based scripts to single-statehandlers (treating all groups in one thread) to reduce scripting overhead. Prior to the patch, I got 18-22 FPS in situations that now result on 12-16  FPS. In the past week, nothing changed in my scripts, but...?

 

 

 

PS. Dear devs: You are probably moving on to the new Engine now, but if you still intend to extend ARMA3s lifetime, here a few suggestions for new funcs for the patches...:

- pathfinding function for buildings: "fnc_findshortestpath  [[room/entry pt start], [room/entry point end]]" => result [array of coordinates or room nodes to be crossed]  (a engine-based solution faster than a scripted Dykstra ...). Moreover, pathfing in hotel or other larger structures is often  poor, AI only manages if you give it step by step only adjacend "rooms" until arrived. Would be good to get this more robust.

 

-> Some improvements to vehicle handling and scripted control:

- fnc_bboxes_will_collide [[entity or config A, presumed position of A, presumed orientation of A], [entity or config B, presumed position of B, presumed orientation of B]] ==> true/false  (again, faster engine based solution would be nice)

- fnc_follow_group [[lead group], [following group(s)], distance, direction]  => Linking two or more groups to a lead group with offset (avoid having to join all units...). And corresponding command to unlink.

 

->Some funcs to influence driving behavior more directly. I see them often veering off the road on the Fallujah map, no matter what combatmode, behaviour, forcefollowroad, engagament status, etc. Oftentimes they crash into each other in column, overtake wildely and chaos ensues. When they crash at odd angles, they sometimes fling across the map or dive below ATL=0. Anything to improve driving or help scriptes to affect AI driving would be welcome.

Any of that would be most appreciated...

[solargazr 1]

6 hours ago, Jan-H said:

Anyone else noticing:

 

(b) 10-20% loss of performance in large, vehicle rich scenarios even with -nologs (due to the first)?

 

Hi, I noticed about 15% drop with latest dev update, it's regardless of scenario/benchmark... even with empty map there's noticeable fps reduction in dev vs stable. Previously dev branch have been equal to stable re performance. Now for example 'Benchmark Altis' from steam workshop gives 68fps average in stable and 58fps in dev.

[solargazr 1]

I did further experimenting and noticed the latest dev build uses about 15% less CPU than the stable build. Could explain the drop in performance if it's not utilizing CPU properly.

[solargazr 1]

The performance drop is also in the latest 'performance' builds for 2.06. The RC for 2.08 does not have this regression, so hopefully next stable update will be smooth.