#perf_prof_branch
1 messages · Page 36 of 1
1.92.145773 new PROFILING branch with PERFORMANCE binaries, v10, server and client, windows 32/64-bit, included linux 32-bit server
- fixed crash in dynamic simulation
My last unit and now my new unit absolutely HATE this stupid idle timer for our small milsim groups. Whats the current news on how to disable it? Im really surprised there isnt more uproar about it because its one of those features that Im sure 99% of server owners never aske...
Hi guys. There was an issue with the profiling branch where smokes would not be visible by users of the prof branch, where as main users could see. Was this the problem with 145755 / 14542 and fixed by 1.92.145767 ?
crashing 🤣
q, anyone here runs the recent profiling branch (performance) binaries and got some unreported crashes ? aka looking for crashdumps
All clear so far
no problems
I have received a crash, with the error message "you don't have a Bohemia Community Wiki account, wink wink nudge nudge"
No chrash but with last perf exe I have this error in few Showcases missions 17:41:57 Error in expression <space getVariable "RscStatic_display")) || !(isNil "BIS_fnc_establishingShot_ski> 17:41:57 Error position: <|| !(isNil "BIS_fnc_establishingShot_ski> 17:41:57 Error Type Bool, expected Bool 17:41:57 File A3\functions_f\GUI\fn_establishingShot.sqf [BIS_fnc_establishingShot], line 264 17:41:57 Error in expression <space getVariable "RscStatic_display") || !(isNil "BIS_fnc_establishingShot_ski> 17:41:57 Error position: <|| !(isNil "BIS_fnc_establishingShot_ski> 17:41:57 Error Type Bool, expected Bool 17:41:57 File A3\functions_f\GUI\fn_establishingShot.sqf [BIS_fnc_establishingShot], line 265
@thin wyvern what is the version of that last exe?
AFAIK this was fixed 1 month ago, why you are running old code, no idea
1.92.145.773
That's just waitUntil thing
no profiling for the weekend, ran into some troubles merging too many fixes , will try get one early next week
skipping v11 to v13 (due to various issues)
1.92.145888 new PROFILING branch with PERFORMANCE binaries, v14, server and client, windows 32/64-bit, included linux 32-bit server
- resolved a lot of crashes (e.g. audio, AI related etc.)
- improved command-line arguments handling (e.g. for linux dedicated server)
- extended #monitor , #monitords commands with more details (networking, players)
- for players info L = lobby , R = role selection , B = briefing , G = in-game , D = debriefing
- for new traffic info NG = non guaranteed , G = guaranteed (?-BE means BattlEye traffic) - fixed debriefing after missionEnd
- when using server.cfg settings (missionsToServerRestart,missionsToServerShutdown)
- when using admin command #shutdownaftermission #restartserveraftermission - fix for freeze related to SAM/LOAL missiles
- script caching is back with more performance and no memory leaks
My last unit and now my new unit absolutely HATE this stupid idle timer for our small milsim groups. Whats the current news on how to disable it? Im really surprised there isnt more uproar about it because its one of those features that Im sure 99% of server owners never aske...
@empty goblet what does all the other stuff mean?
it's right above your post in the changelog
oh, you have edited )
no it was there since first posted, no edit
remember #monitords writes into server console (and it's log) so in that case you don't need use #monitor
@nocturne obsidian it has been fixed by one of the top programmers so...
soo......
@empty goblet we have a problem :D
https://s.sqf.ovh/2019-07-16_19-58-57.mp4
🤦
And nope, it doesn't clean up.
Even after going from editor preview back to main menu, starting new editor preview. Or joining a MP mission, memory doesn't clear anymore till game exit.
But. I honestly have to say. Thanks for giving us a seperate compile cache just for SQS scripts. That was definitely worth the effort and all the SQS scripters will love you for that.
If you need some help implementing it properly, give me a call. I can free some hours for that..

It works as intended up until 15 seconds then it doesn’t lol
There is a bug somewhere
Nah that might be a unrelated memory flush, or the grow of the hashtable to more buckets, but that wouldn't cause such a hard dip
so is that now live on all official servers?
the memory leak in that edge case is wip
there is bug with the #monitords retaining the value after instance autorestart (fails to restart) , also wip
some issue with the commandline parsing (again sideffect vs linux negative) wip
noticed the #monitords reporting negative memory values, also wip
I think we didn't get any details yet as to what the commandline parsing changes were?
there is new profiling already on steam, has fixes related to attachTo vs objects and especially simpleObjects, causing stuttering
@whole cloud those were to address that linux issue vs mods with spaces
also in the new profiling one of the fixes may make the game runs smoother / better vs any objects movement
sideeffect of the fixes
Ah I think I remember. Nice. Thanks o7
1.92.145899 new PROFILING branch with PERFORMANCE binaries, v15, server and client, windows 32/64-bit, included linux 32-bit server
- fixed issues with attachTo objects and simpleObjects causing stutter
- possible performance / smoothness improvement related to moving objects (due to fixes from previous line)
known issues (wip to resolve):
- the memory leak in caching as edge case
- bug with the #monitords retaining the value after instance autorestart (fails to restart)
- issue with the commandline parsing (side-ffect vs linux )
- the #monitords reporting negative memory values
My last unit and now my new unit absolutely HATE this stupid idle timer for our small milsim groups. Whats the current news on how to disable it? Im really surprised there isnt more uproar about it because its one of those features that Im sure 99% of server owners never aske...
waiting for caching to be removed or fixed, before installing
Hey i need help with arma 3 i have a gtx 1060 6 gb and a ryzen 5 1600 but i get no more than 40 fps when i auto detect it put everything on ultra
@inland dew the leak should be only edge case not normal usage
depends on your issue of "normal usage"
ALiVE had so much bad code that your game wouldn't be able to run longer than a few hours without crashing
what is the case, for example?
v14 run for 12 hrs before I've seen the report about leaks and replaced it back by v10
if you compile many different things.
Every script you ever compile stays in cache. If you run mods/missions that compile alot of different things, and especially dynamic content then you'll run into problems
Found that out by the number of complaints of freezing. Rolling back as well.
do you do call compile a lot?
server side, yes
v16 inbound, all but leak fixed from the known issues
let me correct myself, the monitords value is still borked but at least it's now restarting the process
Just tried the new binary on steam, and am getting mod dependencies fails followed by a CTD. Rolled back to the previous version with exactly the same launcher options and mods and all is working
Another player, used the latest perf build on Steam. His game also failed to load all the required mods. Rolled back to previous version and all was good.
So there is an issue with the current perf build on Steam that prevents mods being loaded
I think your looking for #dev_rc_branch
edited ... was referring to the perf build
@trim ether please send me exactly what OS the server / client has and also exact commandline used
@empty goblet It'll not be until later tonight, as I'm at work. However, we did not even get near the server. The mods were not loading correctly even on the main splash screen
ye i but i need the samples to figure out what's wrong ideally with compare vs previous profiling
anyway ye, something is wrong with 145913 , just now figure out what
wait, i didn't put up the 145913 for manual download, so ....
so 145899 is broken too
I've asked someone to run a test over lunchtime, so will have something for you early afternoon
my server test (where i got servermod) 149830 ok, 145888 ok, 145899 fails init mission, 145913 fails too
but no idea how you 'determined' the mod loading failure, cause i see the mod in the list of loaded addons
We were loading 10-15 mods and only one loaded successfully. The remainder did not even show on the splash screen
would be nice to know which one loaded and compare to the rest (name/path)
ok, anyone else who got problems ? i need the exact commandline used and which OS
145913 loads for me on server 2016
it may load, but something is terribly broken for the cases where it is
thus massive revert to ~145888 state (w/o caching)
also @heady parcel it maybe worked cause you had no createSimpleObject in the mission
anyway @trim ether the issue with the commandline was discovered, it was truncanting any similar commandlines and thus anything similar but first 'vanished'
Hmm, but I do. My exile base parts are converted to simple objects.
Ok, thanks for the update
@heady parcel in that case it's even more confusing why it broke only some servers/clients
anyway those problems encountered past 145888 mean that related changes can't be merged into 1.94 stable main (too risky)
also i can't warrant how easy will be profiling sync vs 1.94 to get 1.94.profiling (could be more headache than expected)
but we will try something in two weeks (indefinitive 🤣 )
@empty goblet are you planning on rolling back the version on Steam to 145888 to prevent others from having the same issues ?
145918 is prepared now, it will be at level of 145886 changes (so pre 145888) but i need test it first cause for official servers anything past 145830 was unusable
i think about putting for download the 145916 so you can experiment with but it has that nasty bug where paths with spaces or multiple mods are broken
so it will work only with single mod , (maybe unique directory paths of mods) and short full and any paths only (both to binary and to mod)
mumbles something about how small simple broken everything
😃
it's sad, all those changes were really very promising (w/o the bugs ofc)
2 steps forward, 1 backwards ..... aways the way in s/w dev
145918 now on steam
1.92.145918 new PROFILING branch with PERFORMANCE binaries, v18, server and client, windows 32/64-bit, included linux 32-bit server
- reverted fixes back to ~145886 (script caching, object stutter, known issue of failure to auto-restart instance, linux commandline vs mods)
My last unit and now my new unit absolutely HATE this stupid idle timer for our small milsim groups. Whats the current news on how to disable it? Im really surprised there isnt more uproar about it because its one of those features that Im sure 99% of server owners never aske...
no eta on when script caching , stuter fixes gunna be resolved, rest is already internally done
so caching disabled and no smoothening?
yep
Stupid question here, I haven't read up on these things yet but I'm excited theres still dedication to better performance!
But anyhow, this script caching, is this useful for servers or clients? Basically what does it speed up
No point discussing it now it might have gone for good now
it's useful for everything, and it speeds up stupid code that you should never write anyway
said stupid code will cause memory leak with caching enabled. (unless there is some cache clearing/invalidation added)
they tried adding that, but failed. Which is why it was removed again
did the cache have any limit at all? 
I suppose it's different from the "load mission to memory" option?
Still not sure if that's for speeding up client downloads
yes, cache(s) was used for sqf/sqs
does one have to opt-out perf branch to have full 1.94 until a new version hits?
i assume yes
7.4 GB 😐
i will try get the sync done tomorrow but no warranty, more likely next week
and the bug mess (known issues) declutter after that
hotfix also next week then right?
https://feedback.bistudio.com/T142645 seems like the already fixed headless idle kick is back
ye, sadly quite some of the last group of fixes in 1.92.profiling aren't in 1.94.main (and more likely will be missing from perf0 for 1.94 sync)
hopefully they'll make it into hotfix then
oops
hey, I'm playing my arma 3 on Performance profile Build (Dwarden official) 192perf - perf18. (1.92.145918)
it Really help on stutters on Livonia Exile but I have issues with constant Desyncs/Rubberbanding.. anything I can do on my end to fix that?
I don't want to revert back to normal after that smooth experience but.. desyncs make it unplayable
Just got an update Steam "Arma 3 Server (profiling)" now the server crashes on start
Exception code: C0000005 ACCESS_VIOLATION at 137D525E
Grab the crash report (rpt,bidmp,mdmp) and put in zip file and send dwarden
done
known, the merge isn't complete yet
why is it live then?
Oh Ok I know that issue ^^
_just like in any action comics _
Do you have an ETA for a working build or should I go back to standard for the time beeing
switch back to stable 1.94.main branch for now, atm. no eta
ok thanks 😀
Not really perf prof.
But.. Has anyone ever tried the -interpolation= start parameter? 🤔
has values 0,1,2
0 is "minimal non-interpolated"
1 is "Automatic interpolated"
2 is "Precise Interpolated"
🤔 I didn't know that parameter existed until now
🤔
@whole cloud maybe remnant from A2 network tweaking days?
No idea where from. But it seems useful.
1 seems to be default. so seems like you can increase interpolation
I wonder if that might have a performance impact. As I often hear that interpolation is expensive in MP with many units
@rare river there is new profiling, this time it shall not crash
Thanks for the heads up will check that tomorrow 😃
No crash on start 👌
Guys, forgive me because I am going to sound super stupid but I've always wanted to ask this...
Why isn't the Performance profile Build Binaries the default Arma 3 binaries?
Because these changes are not tested so much, they can break your game sometimes.
It's branch for testing.
After the changes in this branch are fully tested they end up in the base game.
But... as soon as it is tested and proven bug free do they make it into the regular?
Ohhh you answered that!
Yeah.
That is great to know!
The branch is if you want more performance but don't fear occasional problems.
Because this is always SO much faster than the regular binaries I thought they never made it into the default ones but if they do they I can rest in peace now
And when this "merge" occurs is it stated on the forums or somewhere or it just happens in the background? (Like silently)
I mean when the Performance profile Build binaries become the Default Arma 3 binaries... (Are the default Arma Binaries called Profiling Branch?)
no, that's not how it usually works
profiling binaries are eventually updated with any updates to the main binary
so not all profiling binary changes makes it into main binary updates
But the important part... do the parts of the profiling binary that make the game run better DO make it into the binary updates?
Because I am trying to understand if I should be using these performance binaries for my regular play or just wait a bit for the performance parity with the default binaries happens... (It it happens!?)
maybe they do, maybe they don't 😉
either you're on the generally more stable main binary
or you feel the rush with the more experimental performance binaries
sounds good, I prefer speed... too bad the stable guys won't get it and Arma could use all the performance we can get out of everything 😄
@silk summit perf binaries are basically a dev branch, but compatible with the current stable version.
Thus BI can get many more people to test.
performance build doesn't mean more performance btw. It MIGHT contains performance improvements, but after they've been tested they are merged into the base game.
What I needed to hear and that makes the most sense to do, who would lock away performance improvements from the bulk of their audience? Thank you @whole cloud
The problem is that data doesn't change and some exe changes are often connected with changes in data
today no update of profiling , failure on semi auto publishing due to binaries numbering issues
@silk summit when it works it will be given to everyone in upcoming main branch update
v1 profiling 1.94.145968 is on steam/dropbox/google drive btw. but it's not exactly what i wanted to roll out so rest of merges must wait till monday
there missing some fixes for the paths issues, the monitords, autorestarting related stuff and some tweaks
what are the fixes/improvements included a what's missing?
if something behaves worse than in 1.92.profiling or flat out is missing please drop me DM or mention it in the #perf_prof_branch
think will wait for monday )
Good to hear some pathing fixes are going in at some point when it works. Is that just to fix the move to issues or also the AI ramming into trees with vehicles and such?
if on client FPS went up by as much as it does, when you assign more threads to the server...
waitUntilundef expression handling is now error only (but beware , if you get the error, then your script code most likely ends in infinite loops)
- fix for some forgotten array resizes in
getResolution
Oh it contains that thing. OK avoiding that like the plague then. Going to take a while for the community to make the massive code changes to all the mods for that to be workable.
getFuelCargo, getRepairCargo, getAmmoCargowill return-1when the vehicle is not capable of using cargo/repair/ammo
and rest are the 1.92.profiling merges into 1.94.profiling
don't understand very well, caching will be reenabled?
not yet, don't expect that soonish (more like later before 1.96)
thanks god!
unless that code and the other optimization is 'safe' enough i don't want to see it either
Quite a few scaries on the horizon that will be game breaking. Not looking forward to when these get rolled out at all. All that abandoned code we use which we might not even be allowed to edit, not good.
actually only those 2 were problem, the script caching simply because it was missing more of sanity checks
and the other one was rolled too early (too experimental for experimenting) , that was the one related to object visual states / stutter
because restaring the server more often affects online directly. Sure, players that needed to leave anyways, will use server restart to finally leave. But without these restarts they could have spent some more time on it )
that's why i need that build with fixed #monitords vs restarts and endMission vs debriefing when using missionsToServerRestart/Shutdown etc.
what was supposed to be the one which failed to publish 🤣
not useful for our usage )
server gets restarted every 6 hours and also restarted if mission is finished
also i would like to know if the idleKick / timeout stuff is working or broken on the profiling cause it was in the 1.94.main (or some claimed it is)
@inland dew that's what i mean it would be usable for you
it would auto-restart each time it goes to missionEnd after debriefing timeOut expires (if you set the MtSR value to 1)
there is no timer ofc, so that you need to do thru the rcon
it would be useful for other server owners. Since we've already implemented our own solution a long time ago to this and don't require it to be done on arma level. Would be waste of time to recode everything to use arma stuff instead of what we have now to do it
you can always test it and compare to your own solution and DM me the result / your opinion 😉
I think the person who's responsable for this won't be wanting to redo everything since what we have makes what is needed
@nocturne obsidian Oh it contains that thing. OK avoiding that like the plague then. you mean the waitUntil thing?
i would like to know what he means too cause if it's that, then that thing is in all builds 😃
the only difference is what happens when engine now encounter the wrong written script code
yeah the waitUntil thing is currently in public build in it's bad form. The perf/prof build is way better.
So I don't understand why that's bad?
1.94.145986 new PROFILING branch with PERFORMANCE binaries, v02, server and client, windows 32/64-bit, included linux 32-bit server
- sync with 1.94.main
- merge of all 1.92.profiling fixes which aren't in 1.94.main (dozens)
- waitUntil undef expression handling is now error only (instead of break on error)
(beware that if you get the error, then your script code most likely ends in infinite loop) - getResolution extended with topFOV, leftFOV, tripleHead, added forgotten array resizes
- getFuelCargo, getRepairCargo, getAmmoCargo will return -1 when the vehicle is not capable of using cargo/repair/ammo
- fix for #monitor / #monitords to work correctly with missionsToServerRestart
- improved work with command-line arguments (including for linux)
My last unit and now my new unit absolutely HATE this stupid idle timer for our small milsim groups. Whats the current news on how to disable it? Im really surprised there isnt more uproar about it because its one of those features that Im sure 99% of server owners never aske...
@empty goblet I guess the thread name should be changed from 1.92 to 1.94 😉
Linux has newer buildnumber? that correct?
ye it got built later at different loop so higher version
@rain moth semantics 😉 number changed
now this will be installed!
i have the same error @inland dew
When trying to log on to any official server after 1.94 hotfix, the error message shows up: Session lost - Verifying PBO failed enoch\addons\functions_f_enoch.ebo Update was forced by steam, data files were validated after update.
Ive forwarded it.
It just means that particular server hasn't updated. That's what shows if you're on latest client version and server is on 1.94 vanilla
dozens of people get that issue, and they say they have updated their server
So it's not just the official servers @whole cloud?
@inland dew Did you do a full update?
I updated two of my servers (and my client) @jaunty kestrel No issues
Updating the last one now.
Make sure that people update the client ARMA on their PCs ... a lot of time they forget that they need to leave the game 😉
Good to hear 😉
@hallow yew That error also happens if the client is not updated but the server is
@obsidian condor same error if client updated but server not
Yeah, was just clarifying The Tall Man's earlier statement
A real mess.
- Yesterday perf build and already today not usable anymore.
- Hotfix released to soon. Updates and hotfixes should be released early afternoon european time so US server owners are awake and it's not to late for eastern countries.
- Lack of communication at BI (I suppose), because they were reminded by regular players to update their official servers.
@hallow yew hotfix build number is lower than yesterday's perf build
Yep, my mistake, perfv02 was 145986 Windows and 145991 Linux
@inland dew data changed that's why people can't join , unrelated to profiling build
both 145968 and 145986 works as server with 145777 data for clients with updated 145977 data
I understand, but then what was the point to push v01 las t week and v02 yesterday?
would have been better to directly push perf build like tomorrow or don't know
so after the hotfix
as i explained at friday, the publishing part of the v02 failed, otherwise it would be out at friday
and profiling missing only contact related fix merges, so it works fine vs hotfix clients
new profiling was already built today as sync, but again some trouble prevents publishing it 😉
and as also explained before profiling is way ahead , even of hotfix ...
what I mean is wouldn't it have been better to wait for hotfix, as it was coming anyways and then push first profiling
why wait ? the profiling was already done last week, sigh
don't look at build numbers, they unrelated between branches
so, now the main, profiling and creatordlc server base data branches should be 1:1
ofc creatordlc has the gm stuff and profiling has own binaries 😉
Yesterday perf build and already today not usable anymore.
Why not usable anymore?? Noone said that. It works just fine.
it was unuable when data were different 😉 just he didn't knew
so, if there are any problems just write it
No build = no problems
what no build? there is build, it's the 1.94.145986
it's ahead of main (unless you want to play contact on that build)
all the official servers are already on 145986
If i use the Client Performance... how can i solve the problem that it uses a new player profile?
its "name"
justice , you use the steam client to switch branches or ran the .exe from manual download ?
i run the exe from the manual download
if you just replaced (backup old) the arma3_x64.exe then it should use same profile
unless your profile has some space/weird characters in it (or path to it)
it has a space in it
but first try just replace the default main binary with profiling branch performance binary
Justice Molton
kk
so usually my name is Justice Molton and with that changed .exe its "Justice Molton"
@onyx granite could be related to the changes to fix issues with spaces and such (in paths, in profile names and mod names etc.)
@inland dew
No build = no problems
We have perf/prof v2. That is a build.
So I heard about MUCH better performance on livonia on dev-branch. These fixes apparently didn't land in hotfix, are there any details about what has been fixed? Engine or data fixed on Livonia? And why they were not in hotfix? Livonia performance is the main thing people complain about, so it would've seemed sensible to fix that in hotfix, if it's already fixed on dev branch..
Also the waitUntil thing trying to fix infinite loops on undefined variables.
Now people just use getVariable with a default value instead, and the waitUntil's will still loop indefinitely. I really don't see any purpose in that "fix"
Before: waitUntil {(player getvariable "loaded") isEqualTo 1}; loaded never gets defined == infinite loop.
Now: Error and script dies.
Scripter updates his script to
waitUntil {(player getvariable ["loaded", 0]) isEqualTo 1};
End result, loaded never gets defined == infinite loop.
Total benefit: minus one. Scripters need to update their code, and still have infinite loops.
Or same with
waitUntil {MyVariable == 5} MyVariable never defined == infinite loop.
Fixed version:
waitUntil {!isNil "myVariable" && {MyVariable == 5}} MyVariable never defined == infinite loop.
Benefit? Uh... More code, more effort, same outcome.. 🤔
Yeah I don't see the benefit. It doesn't feel like a fix at all.
Especially at this age of the game.
infinite loops on undefined variables
I don't get how that is the issue.
Couldn't someone do:
waitUntil {my_global}; // global is nil
and change my_global to true at some point to stop the waiting?
Yes. Sure. The issue is that when people never set the variable, it will loop indefinitely
that is "fixed" now. But instead people will now add isNil checks or default values, and they will still loop indefinitely
That is my point.
That change looks stuping and like a change of sake for changing stuff.
Same here. It's completely useless, besides forcing people to revisit 10 year old code
If having nil working as a "false" was fine for 6 years
then suddenly change like that is just a stupid BC break for no reason at all.
You just don't change public APIs like that.
If you've fked up and had stupid behaviour allowed in your production api that was widely adopted you just swallow that or introduce api v2
You don't break code of all your "customers" because you don't like that old code after six years.
¯_(ツ)_/¯
3rd
why not just log to rpt?
It logs to rpt right now. But it's planned to break that in future again.
better fix later than never
BI: We strongly recommend fixing scripts however, because we intend to force exiting the loop in 1.96
what should be the benefits?
more/better performance?
Couldn't someone do:
they can't without having rpt filled with undefined error in seconds. And just before you say something like "you see what they have done" this was the case forever. With the break this loop would have stopped after 1st error
Undefined variable will error regardless and will log error at the rate of waitUntil. But getVariable does not error when you retrieve non existent variable unless you assign the result to something and if it is in expression the whole expression is nil with no error
not to mention millions of whaitUntil {if (something) exitWith {}; /carry on/}; another script in the scheduler that mightnever exit but will be put up for evaluation every simulation.
not to mention millions of whaitUntil {if (something) exitWith {}; /carry on/};
And what's the difference if someone will do whaitUntil {if (something) exitWith {false}; /carry on/}; after the "fix" ?
None.
It's still the reposnsiblity of the scripter to write something in a way that won't be stuck forever. I won't repeat myself. Read what I wrote earlier, this is just a BC break for no reason at all.
they can't without having rpt filled with undefined error in seconds.
Isn't it currently throwing the warning once to RPT (as suggested by Dedmen)? It's supposed to be some sort of deprecation warning, however surely noone will care about a deprecation warning for code that it's old and unmaintaned but works. Stuff will just break.
And what's the difference if someone will do
the difference it will not do what they wanted. What they wanted is to break from the loop with that hence they need to set it to {true}
because in majority of cases this is what the intention was
Isn't it currently throwing the warning once to RPT (as suggested by Dedmen)
it currently throws undefined variable error regardless at the rate of waitUntil
this hasnt changed
the new error handling is for expression evaluation not undefined variable detection
Is it just me or does it seem like a bad practice to run a loop unchecked at all?
there is new profiling 1.94.145999 (already on steam branch) today fixing the "" issue with profiles/names
(aka old profiles work again correctly (no new folder made) and new ones too)
1.94.145999 new PROFILING branch with PERFORMANCE binaries, v03, server and client, windows 32/64-bit, included linux 32-bit server
- fixed the
quotation marksissue for operations with profiles/names
(aka old profiles work again correctly (no new folder made) and same applies to new profiles)
My last unit and now my new unit absolutely HATE this stupid idle timer for our small milsim groups. Whats the current news on how to disable it? Im really surprised there isnt more uproar about it because its one of those features that Im sure 99% of server owners never aske...
When I try to use the server performance binary, I get "battleye initialization failed" error in the server console. Anyone else have this issue?
oof. Apparently perf/prof didn't fix my crashing issues :/
Who you got a crash on?
Friend had around 5 crashes yesterday. Every time his RPT ended with same thing.
22:25:12 Loading movesType CfgMovesBird
22:25:12 Reading cached action map data
@empty goblet 1.94.145999 killed Arma3Profile config on one of the servers (all others are fine)
with 1.94.145999, each time each player dies, in the chat it's displayed twice, like this:
Groove_C was killed
Groove_C was killed
despite me dying only once at a time
Okey no crashes anymore besides that one.
Will have to test further next week i guess
@limpid crane explain killed profile config ?
@whole cloud remember to send me crashdumps after weekend from last build
I think my crashdump might be nvidia driver fault
that applies to anyone else, so it can be fixed sooner than later
other people already told me latest driver is bad
nvidia has tons of security and stability exploits, they trying to fix it but the CVEs are pilling up faster than new drivers
so i'm not surprised
@empty goblet I will tell you the story instead of MiB.
After server restart with new perf build, profile was there, but it was empty.
which file exactly ? or whole folder ?
folder and file were there
but inside of the file were only some SP related settings, like VON, music etc
nothing from 3rd person, skill, precision, camerashake, statistics, visibility of nicknames of players talking, distance and direction to objective etc were not there
such stuff never happened to our servers before, since A1 times
I think this was the only stuff left/present in the profile:
maxSamplesPlayed=96;
sceneComplexity=400000;
shadowZDistance=100;
viewDistance=1000;
preferredObjectViewDistance=800;
terrainGrid=25;
volumeCD=10;
volumeFX=10;
volumeSpeech=10;
volumeVoN=10;
vonRecThreshold=0.029999999;
and I think that the server run at expert, despite the fact that there was written "custom", because players have reported thirdPersonView 0, scoreTable 0, vonID 0, waypoints 0 etc.
don't know if profile was somehow deleted and automatically recreated, partially or what happened
our KotH servers and AAS server haven't suffered from this problem with 1.94.145999, strangely
not sure if it helps but, I'm running perf v3 and have been for about 12 hours which makes about 4 restarts on 3 of my servers; one a3 epoch and 2 exile. No crashes or problems with profile. All running as expected.
Will post here again if that status changes
@inland dew anything (errors) in client/server RPT ?
@empty goblet I found the problem with display twice kill message in chat. We have our own system which dislpays this message.
That's why we had to disable the default one in the profile. But it was reenabled, after we had to put back default profile after profile loss with update to 1.94.145999
so, how is 1.94.145999 ? any crashes (i'm aware of some rare ones, so if anyone has some crashdumps please dm and share)
Me happy so far
Same here, fine on servers and client
No problems
All good here too
Good everything is fine. Means you can now enable the next level of MOAR FPS switches
wait! have seen very low FPS on server end, despite it being normal on both HCs
need to see waht it could be
server frequency is stable - no throttling
yestreday I've seen lass than 20 FPS on server around 23:00, with restart happened at 17:00
normally it's much more than that even 12 hrs after restart, when next resart comes
but not that low and not just 5-6 hrs after restart
since 145999 installation I haven't checked FPS until yesterday
just that it's not crashing or anything esle, I can confirm
have even checked FPS this morning after restart at 05:00 and until next restart at 11:00
we were not even 10 players and once again, servert FPS was sub 20, like not even 5-6 hours after restart, whereas usually it never goes so low with server full, even after 12 hrs
will ask once again for server CPU frequency, if it didn't dropped for some reason
need may be to switch back to stable and if on stable the problem is not present, then it's something wrong with 145999
don't as k me what )
have checked random rpts starting from 9th August with 145999 and FPS wise, server was for the most of the time like on stable 1.94, after 12 hrs
need a few days more to compare
need confirmation if server CPU frequency was stable and if yes, tomorrow morning I will see again server FPS after 12 hrs, starting from today 17:00
if low, will switch to stable and let it run for 12 hrs as well
@empty goblet FPS between 145977 stable and 145999 perf seams to be +- same.
2x 12 hours on 145977 vs 2x 12 hours on 145999 perf
ye there is no reason it to be much different, it's mostly bug fixes and some features from the past profiling
Will move back to 145999
We have experienced two occurrences of a new bug to us since Contact.
Players: 2 and 6
Server load: fairly light in both cases. No combat occurring
Time: 10 mins from start and 4 hours from start
Build: vanilla 1.94 and perf 1.94.145999
Issue: Clients lose all data sync with the server. Each client sees their own divergent view of the game, but are completely unaware that they are not talking to the server or each other. FPS on both the clients and servers remains stable. There is nothing in either the server or client RPTs. If a player DCs and they are unable to reconnect.
As I said above, this is the first time we have experienced anything like this.
The only reason that I'm posting it here as it involves the perf build.
If a player DCs and they are unable to reconnect.
that in itself feels like the rare bug introduced with the change adding BattlEye's own data port
i've theory about what is happening but there never was any reliable repro
server restart or client Ip change fixes it
but losing touch with game w/o any disconect at either end, that's new
@empty goblet We do not run BattleEye on any of our servers
well then no idea, need reliable repro or indepth details to track such rare problems
Understood
Think that be issue cleared sometime back @empty goblet
Haven't seen it for a while
@empty goblet could you eventually tell us (me) what this is?
Fixed: Memory alignment issues
Memoy leak(s) fixed?
Also what does this mean?
Added: #monitor and #monitorDS are now retained after server restarts
Fixed: #monitor and #monitorDS were not transferred after server restart
If I don't logout on server restart and rejoin after restart, I will be automatically logged in and it will have remembered that before restart I was logged in and have entered #monitor 3 for example and will continue to show me the info every 3 secs?
With dedicated server I don't really understand as well.
I know that with DS one can log info serverside, but what about after server restarted?
Memoy leak(s) fixed?
no. memory alignment. Doesn't mean anything to you or for any player, but broke atleast one of my mods.
alignment is something a C/C++ programmer thinks about.
@inland dew when using server setting to restart after N missions, it will remember the previous set value for #monitords
see the changelogs i post in #perf_prof_branch
to those wondering, 1.96.146083 is profiling v0 in sync with 1.96.main (146070) already out since yesterday
any word on fix for broken BE path ?
not yet
ty
@inland dew memory leak doesn't influence performance in any way. Its just a memory leak
no situation that makes windows juggle around more with ram addressing, or moving stuff to swap file?
well it doesn't influence performance until your memory runs out and everything starts crashing.
Need the weekend to test, but first impression is that somehow server FPS degrades even less over time and if instead of 30 players, only 20 or 10 left, FPS is higher again.
Not as high as after restart with same 20 or 10 players, but definitely higher vs. 1.94, where FPS remained same or recovered less with less players left
Hope it's not just placebo
What is the default value for unsafeCVL in the current profiling version? And is that different from stable?
afaik we are on prof v0? thats just same as main branch
Hmmm, interesting.
Because CVL isn't working for me when testing ZEN, but it is for one of the other devs. And the only difference is that I am on the profiling branch and he is on stable. Neither of us set unsafeCVL
Specifically, we use CVL to create a logic to use as a target for a camera.
prof was built before the release build? :3
Finally we are important!
how could you ever think you weren't ...
😍
I would like to see why my linux server performance goes down even after I despawn AIs. Would be cool to have prof build for linux. Where can I get it?
Can't find it in downloads at the BI thread for this version, but I guess I can dl it through steam, I didn't know that before.
Steam branch. Others aren't avail yet
Maybe someone has the profiling x64 .exe? Steam wants to download 20 GB through my slow internet to switch to the prof branch...
Or it won't work anyway?
I do yeah
But I'm sitting in Prague airport rn :D message me in a couple hrs
I mean you are offline rn anyway so whatevs
prof branch not updated with latest GM?
yeah it didn't get update. And prolly wont soon
that's too bad, makes it a lot more annoying to switch between 
That made me remember sparker asked about it 😄
just looking at the steam metadata, why was multiplayer removed from categories? 🤔
profiling CDLC depot should be updated since 2 days ago
which build number (file version) is your prof exe?
ah no new prof build, but the gm folder in prof was updated
I do yeah
Thanks, I just unsubscribed from GM and I have it now, steam is weird
Is there a list of known section names somewhere?
not really
Not really or totally not? 😄
Okay okay, I've got this
BTW is prof build supposed to run better in any way?
I suppose so 🤔
possible server fps increases and clients may experience higher fps too
no
perf/prof both sometimes have new improvements before the normal build, like dev branch. and new improvements sometimes improve performance
Okay so I will start listing my assumptions about the naming here 🤷♂️
Main thread/total/sLoop/wSimu/rendr/wDraw/perAI/iAll/aiGroup/aiSub/aiUnt/AiPth must be the path finding per individual unit I assume?
It's a dedicated server btw, I am confused to see sections like 'rendr' and 'wDraw', is it normal?
sounds like it yeah
rendr doesn't only do graphics rendering
also like.. display eh's.. which might not fire on server.. uh...
yes normal but.. dunno
Did anyone... maybe... have any issues with the JIP queue in the profiling server branch?
Is JIP queue behavior somehow different in the profiling branch?
99% That my own code is behaving incorrect of course. But maybe there is information that JIP is different in the prof branch, I dunno...
Never mind, found an error in my code, I am an idiot
error guys
CPU doesn't support invariant TSC. Define TRACY_NO_INVARIANT_CHECK = 1 to ignore this error, * if you know what you are doing*
well you should report a bug on my profiler then :D
windows or linux?
and that has nothing to do with #perf_prof_branch
@whole cloud do you know if the get mouse position fix made it to 1.98?

I'll see if we can get it into 1.98, but I think I'd prefer not to, I don't want to have another post-update hotfix..
its always the "simple fix" that breaks everything in the end 😄
Could just bet that we'll need a hotfix either way because someone already merged something broken
Heh. Hope not.
Maybe its best to do the release on the end of the week 😄 Cuz I work on saturdays
@fickle geyser can you find me the FT ticket and link me?
Its not fixed, but I'll make sure it atleast makes it to 1.98, can't promise 1.96, but seems possible
remove the fix link pls :D
Thanks
Thanks
No, thank You for caring 😛

are you all still looking for crash logs in preparation for the next update?
always
preferably from RC branch tho
its a bit late to now report crashes from 6 months ago
yo i use the profiling branch and i get the error message : 0xC000001D - STATUS_ILLEGAL_INSTRUCTION what should i do or what does it mean tho?
what CPU do you have
i5-2500
tried verifying files in steam?
steam says arma is not installed and i dont want to reinstall it.
but it seems like its the problem of the allocator
Which allocator are you trying to use?
cma from blub
There are several versions of it.
AVX or AVX2 version?
how do i know what is what tho?
Well, your CPU doesn't support that. So...
And it has nothing to do with perf branch
And one should first read the instructions, know what CPU is capable of and understand it all.
Not just copy/paste stuff and wonder why it doesn't work
Well on the guide it doesnt said if i should look for it. and it doesnt says it anywhere else.
Hey! Is it normal for arma3profiling_x64 to not be used by the launcher? The launcher just launches the regular arma3_x64.exe.
Thank you
yes
I'm not seeing any difference between using the performance profiling build and the regular one. No difference in fps, frametimes or cpu/gpu usage. Is that normal?
perf build should be equivalent to normal release
profiling should be a bit lower performance
Hmm I tried installing the performance build but I was not able to use the launcher and could not join any servers due to the version being too old
version too old? weird.. its 1.96 it should match
it has a different build number but that shouldn't matter
I might have missed a step then
I've seen a second report of that a day or so ago, but can only explain by serverside missconfiguration
Is there supposed to be a 196perf in the dropbox or am I supposed to use the 194perf one?
no dropbox for this one
only steam till now
I'll try to get a new perf till end of the week, but not very likely
So there is not a 1.96 performance build available, only the profiling build for 1.96?
wondering what will change in next perf
there is 196 perf/prof. performance and profiling are always paired together
both are on the perf/prof steam branch
Ok, can I only use the performance build and not the profiling build from there?
Launcher shows profiling
but you can just rename arma3profiling_x64.exe to arma3_x64.exe
then the launcher will launch that
I'll try that
by default it launches performance build on perf/prof branch
Do I rename my old arma3_x64.exe to something or delete it?
whatever you like ^^
Thank you
A development build-watermark is now showing while playing but I don't really care about it unless it hinders gameplay
Is there anyway to remove the watermark? I can play on servers that use battleye, everything seems to be fine so far.
no way to remove the watermark
after running the Yet Another Arma* 3 Benchmark 3 times my fps has not changed
That is to be expected, profiling build is equal to base Arma 3* with profiling on top, which will make it a bit slower.
So if anything its slower.
Hi.
Is there a dockerized version of server perf branch?
(otherwise, im going to run one to diagnose our fps drops)
you just need to provide the perf branch instead for the steamcmd call
@hollow island When you run the performance branch does it show in the top left corner of the launcher similar as when you run the profiling branch?
there's no dedicated server linux build for performance, as stated by https://dev.arma3.com/dev-branch ? (hence, i wont be able to contribute 😦 )
performance is/was equal to normal stable release anyway
And yes linux performance builds are a thing
1.96.146277 new PROFILING branch with PERFORMANCE binaries, v01, server and client, windows 32/64-bit, included linux 32-bit server
Added: New -limitFPS= start parameter to adjust server FPS limit between 5-1000 FPS (default 50)
Fixed: Chat-related crash
Fixed: Ability to equip vehicle weapons as a player (has been disabled)
Fixed: Ability to move items in the player's inventory while aboard a vehicle (https://feedback.bistudio.com/T81676)
(Currently only available via Steam. Dropbox/gdrive WIP)
you can limit the FPS via nVidia control pannel or with MSI afterburner and many many other ways
Because vsync. I can get steady 45FPS but with vsync you go 60-30-20. I know that Groove but is better if you have that option in game engine.
no, it's not
better to have it on drivers level than in the game, since drivers are better than the game
and disable vsync and choose fast sync instead
And tripleBuffering you must manualy enable in .cfg file...and I'm with red team
well, it's an AMD problem then and not Arma problem
no need for v-sync and not need for triple buffering
Vsinc need t-buffering but it's the problem with your arrogance that you think that you know everything.
?
better ask AMD on their forum(s) to improve their drivers, making it possible to lock FPS on drivers level and to be able to choose something other than v-sync only, like fast sync or adaptive sync
AMD supports free sync / adaptive sync, as long as your monitor does as well
You can do that with AMD card already.
better to ask 1 company to do their job properly than ask devs literally from every possible game to change stuff
Chill is the name.
well, fast sync is not free sync, just so you know
read and see it's the best option to use
so no need for v-sync with high input lag, since there is no tearing with fast sync and much much lower input lag
Arma uses alternate sync and also triple buffering , see the arma3.cfg, tripleBuffering=1; (yep, experimental, non-documented setting)
Fast sync is terrible at framerates lower than 2x the refresh rate, which is usually hard to get in Arma @inland dew . Fast sync is designed for high framerate, mostly e-sports titles. It's definitely not a vsync replacement lol. When running fast sync under 2x your refresh rate you will notice microstutters, increasing as your framerate drops lower...There's a reason the other options still exist. To be fair fastsync is useless since if you're getting atleast 2x your monitors refresh rate tearing is hardly noticeable anymore, and variable rate sync (freesync and gsync) exist nowadays..
@empty goblet What is alternate sync? Is this a customized vsync?
it means it's not typical double buffering sync
I see, it's remarkably smooth though from what I remember. I'm never going back to normal sync solutions though 😄
Shift+ - + FPS is "normal" double buffer or in house,BIS, double buffer? @empty goblet
-limitFPS= start parameter
can be useful on the client for debugging/testing - in other words simulate slower pcs. this is what BI used it for too
if you need testing use the fps limit cheat. Thats already a thing and shouldn't need a param
also in MP?
not sure if its limited to diag binary?
just use whatever gpu driver tool you have to limit fps then ¯_(ツ)_/¯
many possible ways, but I don't think a start parameter is sensible
@whole cloud @feral harness much kudos for your work so far 👏
There is still quite a lot in review. Will try to keep pumping.
I only checked two items off my todo list, which has dozens 😄
Fast sync is terrible at framerates lower than 2x the refresh rate, which is usually hard to get in Arma @inland dew . Fast sync is designed for high framerate, mostly e-sports titles. It's definitely not a vsync replacement lol. When running fast sync under 2x your refresh rate you will notice microstutters, increasing as your framerate drops lower...There's a reason the other options still exist. To be fair fastsync is useless since if you're getting atleast 2x your monitors refresh rate tearing is hardly noticeable anymore, and variable rate sync (freesync and gsync) exist nowadays..
@mellow lark can't say this as I've been using it in Arma for a long time and no tearing or stuttering at all
@inland dew https://youtu.be/L07t_mY2LEU?t=713
In that part of the video you can easily see fast sync affects smoothness of the image
@whole cloud What is new in 1.96.146304 ?
nothing
Thanks 🤷🏼♂️
note that perf02 is now also available for manual download from the usual locations
Thank you, Dwarden
is "-limitFPS" available on stable(not prof) branch?
no
the perf/prof binary is compatible with the stable branch, just replace the binary on the server and stable clients can still connect
yummy!ill do it tomorrow (i dont want to mess with the bandwith test tonight)
🙊
so I hear we need a new perf/prof build 😉 
tomorrow 🤞
looking forward to it!
The only difference with stable is limitFPS
what's the latest prof download for client?
146325 is latest
thanks
i don't find any download for it
i looked at the forums
and the pins here
its on steam branch
its always on steam first
don't think this one will be pushed to dropbox, i expect new build on friday
oh ok, how do i access to the steam branch (its been a long time since i did this)
windows steam or steamcmd or linux steamcmd?
windows steam
Arma 3 -> properties -> beta's -> CautionSpecialProfilingAndTestingBranchArma3 code
146325 <-this code?
ohh ok
i thought the "CautionSpecialProfilingAndTestingBranchArma3" is a placeholder
😄
limitfps on steam branch?
so no 50 fps limit by default?
server binary argument
command line 😉
can confirm, works great on linux 🎉
even a low freq epyc is happily working at 100+ fps in a test CTI mission
-limitfps=<value> is only for Dedicated or also for Hosted?
I'm understand right about a steamcmd command?
app_update 233780 -beta profiling -betapassword CautionSpecialProfilingAndTestingBranchArma3 validate
This is the same binary file for performance and for profiling?
default is performance binary, profiling binary has different name
that's to avoid people run it, unless they need it to actually profile
And original non-perf binary has the same size as a perf binary?
@silk summit are you stuck at 50fps when you host a local server?
And original non-perf binary has the same size as a perf binary?
no, probably different by a few bytes
Linux server binary
bba01dee13a1908abf8f8ade6b5b784e arma3server
$ sha256sum arma3server
0e3c6d54a7a84a4826ef321727411f26182578536a5fd928feb9eae06271de45 arma3server```
as of last profiling update for server depot done at 15 April 2020 – 08:55:31 UTC
@silk summit dedi and HC
no, probably different by a few bytes
@whole cloud hm, then it's a bit strange, because after anapp_update ... validateI have the next:
39660 -rwxr-xr-x 1 steam steam 40611752 Apr 16 02:57 arma3server
39660 -rwxr-xr-x 1 steam steam 40611752 Apr 16 02:05 arma3server.bak
40208 -rwxr-xr-x 1 steam steam 41168968 Apr 16 01:57 arma3serverprofiling
arma3server and arma3server.bak both have the same size
but a different md5
I've asked because I was confused and not sure - was it updated or not. Just wanted to know about a size.
start it and look at the build number
@whole cloud and this is related to my question in #server_linux My Arma server log file started now with a warning
13:42:56 Could not enable linux core dumps. Error 1 - .
and has no any version or build information
hm? sorry, just a moment
the core dump fail is related to your system, not the binary
@silk summit are you stuck at 50fps when you host a local server?
@whole cloud Nope, but did went from 60 Fps average to 52-54... that is why I asked...
you aren't stuck at 50 fps when hosting local server, because that fps limit doesn't apply to you at all.
Thus changing the existing fps limit (which doesn't apply to you), also doesn't change anything for you
Allright that is good to know!
I did mentioned before that even thou I am getting a few less FPSs with this patch, it is much less stuttery and feels better overall than before
I can also feel the improvement in physics handling in multiplayer... so Good Job!
multiplayer physics is broken though :D
And there were no changes which improve performance. Besides the serverside limitfps testing which can cause less stutter
profiling 1.98.146344 out at steam, windows and linux, also on dropbox/google drive ... if all goes well there will be another update tomorrow with some crashfixes
atm. this update may fix some of the physics issues in mp
@empty goblet what's the password for the profiling branch steam depot?
@whole cloud Thank you!
probably a stupid question, but does client have to have perf binary to join perf binary server?
it kicks me out saying something about cur_mp or something
(__cur_mp.pbo) I'm pretty sure
no
Hmm, after updating my server to the perf binaries everytime I try to load my previously working mission it loads tanoa instead
sounds like missing mods
odd, running the perf binary, had a mission with alive running 50+ civilians not including enemies, was running great 60 fps or so with two headless clients, and then all of a sudden fps dropped to 0 and console is printing Overflow repeatedly
Probably object out of bounds. Fallen under the map, or coordinates in excess of 1e5
I wonder if the high frames would make something like that more likely
because I've been stress testing this a lot and haven't been able to make this happen
the only change is the perf binary
I usually get 40 or so fps though
I was getting 120+ at certain points
it was buttery smooth
really impressed with the performance benefit
Really need better logging for what's actually causing the overflow spam.
It's some object somewhere
We had objects falling through ground but we fixed that
Find which object is too far away
allVehicles select {getPosWorld _x contains any elements > 1e5}
Can't script on phone
I will do my best to make that work, not too familiar with the sqf syntax but that is easy enough to understand. Gotta figure out how to enable debug console on dedi for #adminLogged
ahh found it, enableDebugConsole = 1;
does your client RPT also spam overflow? or only server
allMissionObjects "All" select {!((getPosWorld _x select {_x > 1e5 || _x < -1e3}) isEqualTo [])}
That will return all broken objects. Though not sure if that would need to run on server
profiling 1.98.146356 out at steam, windows and linux, also on dropbox/google drive ... physics fix and crashfixes
I'm assuming both server and client are required to be profiling build for the physics fix to fully work?
yes
noot quite. Clientside is more important
buuut.. VERY recommended to have it on server too
spawned by server (vanilla)
both server and client have latest perf/prof?
have asked to update today, but not sure if server already updated
myself stable for the moment
So. You are reporting a bug, in #perf_prof_branch, which is supposidely fixed on perf prof branch. While you are yourself not even on perf prof, and the server probably isn't either
client perf is so vehilces from me appear normal to other players or for vehicles from server/other players to be displayed normally for me?
you need the fix to see vehicles normally
and the server needs it too so that you don't break its vehicles
so if I update client side, I normally will see the correct way vehicles from server and other players?
should
will try client perf now
thought server side was enough
hope there will be a hotfix to stable
because you simply can't explain to everyone where and how to switch to perf
Well, latest perf on client and server (vanilla) and same still occurs
both sides needs to have the latest perf/prof
hey @inland dew stop molesting my boy Dwarden, he is busy with Arma 4!
Lol
https://giphy.com/gifs/13NRZwjF5J17ZS @silk summit
note to all admins
make sure you use kick for disconnectTimeout e.g. kickClientsOnSlowNetwork[] = { 1, 1, 0, 1 };
if the last number is 0 your server may experience random stalls (zero fps) and most/all-player drops
Is this only on perf? What's the default value as the wiki suggests it's set like this:
kickClientsOnSlowNetwork[] = { 0, 0, 0, 0 };
oo err
what you posted
so, if we don't have that value in server.cfg which, I don't, then it will be defaulting to a value that can crash/cause problems?
ok but, only if you're running one or more of these:
{<MaxPing>, <MaxPacketLoss>, <MaxDesync>, <DisconnectTimeout>}
What if you don't have any of those set up in server.cfg?
0 (logging) should be default for all of them
1 as kicking wasn't set to default cause it took some months and many profiling builds to get it work properly
and as i mentioned maxDesync wasn't finalized ever due to that flawed desync value
1.98.146365 new PROFILING branch with PERFORMANCE binaries, v03, server and client, windows 32/64-bit, included linux 32-bit server
Fixed: Second attempt to fix multiplayer vehicle desync
Fixed: Erratic movement when riding jetski in high waves
(Currently only available via Steam. Dropbox/Google Drive WIP)
This should once and for all fix the vehicle desync.
@midnight bison @kindred radish @cold peak
I didn't get any real feedback for the last attempt, please send feedback directly to me. And tell me when it seems to be fixed.
No feedback might delay a potential hotfix
This should be installed on clients, so you would need some of your players to install it.
Installing it on the server will have some effect, especially concerning saving vehicle positions to database.
But if only installed on server, clients may still see desync
Any link for the builds ? Can't find the 1.98.146365
Can i just throw it on the live server and test it, and everyone else with the normal build can still join ? because testing it local probably doesnt make sense 🤔
clients will need the latest binary from the profiling steam branch
to get all the fixes
normal binary can still join as well
not sure if that impacts the testing though, I'll leave that to @whole cloud 😄
allright, i will try it on next server restart and get some clients to test it ^^
they can still join yes
but for them to see the effect, they should be running profiling branch. Its a clientside fix really
its just the server exe thats needs to be replaced ?
allright, thanks 🙂
does it have to be the arma3serverprofiling_x64.exe or can i use the arma3server_x64.exe aswell ?
both got updated
doesn't matter which
thought so, just to be sure
if you don't need profiling, use the normal one
the default binaries on profiling branch are the performance ones right?
yes
I'm always overwriting them with profling ones so I can launch it via launcher 😄
just use the profiling branch and you will have those automatically 😉
hi ! I'm using the client perf branch since one month and noticed that the 1.96 brought me a 20 fps boost so i was playing with 100-110 fps all the time. Since the new 1.98 version came out i've lost this fps bonus is it normal ? thank u for ur answer
@whole cloud
Have u got any idea @empty goblet ?
performance branch didn't have any performance improvements. neither 1.96 nor 1.98
problem still not solved with vehicles floating in the air, with server and client on latest perf
What
But you just said it was good 😢
Any other clients on the server, not on latest perf?
it was good when I joined, but after some players used free vehicles at base, when they respawned, it was same as with previous perf builds
most of clients are on stable
but I'm on latest perf as client (server as well)
Don't know why it happens to Hunters
in our mission, vehicles when spawned are dropped from a small altitude of few meters to avoid them exploding when player gets in, like it was the case when they were spawning on the ground and were exploding if the ground was uneven and a little part of the vehicle could be in ground texture
we have hundreds of players daily and I can't tell to each of them to switch to perf and also to explain to each of them how and where to do it (((
Now I see that +- half of free vehicles have spawned and froze in the air, the other half are on the ground, like it should be
so it's not always and not for every vehicle, despite the fact that this are exactly same vehicles
vehicle spawning a bit floating might also be a problem with your spawn code
maybe, I'm not the one who has written it
but it work just fine since Jan 2018 until recently
how are vehicles spawned? on clientside or on server?
can't tell, since I don't know. must ask
Okey so.
To explain what I see now.
You report our fix as not working.
Because your spawn script, probably running on clients that don't even have the fix installed, is spawning vehicles a bit above ground.
Didn't I tell you to stop wasting my time?
Well, vehicles you see on screenshot(s) are spawned by mission, not spawned/bought by players
I think they're spawned on server
But thats what I just asked
I think he may be getting at: Get the answers and come back with them, because I can't answer anything concretely without them.
because when some vehicles were sinking a little bit in the ground, before it got fixed in one of the patches, it was visible like this for me only on vehicles that were bought by an other player, but as soon as I got in, they were not a little bit in the ground anymore
So I'm almost 100% sure, that free vehicles spawned at main base by the mission, are spawn on server side and vehicles that players buy, are spawn on their machines
So the behavior you see on my screens occurs for vehicles spawned on server, patched to the latest perf build and me, observing this, I also have latest perf build on my client, so it's all correct then
If the way the script to spawn vehicles works, was not correct, we would have had this problem since forever and not only since last patch?
What we want to know is the vehicle desync. Your vehicle spawn script that spawns vehicles a bit above ground, resulting in vehicles being spawned a bit above ground... is not what we are testing
how are vehicles spawned? on clientside or on server?
@whole cloud The MRAPs on the screenshot are server-spawned
to me looks like they are just not physx simulated, which is not the bug that we were trying to fix
We are trying to see if the critical bug of vehicle desyncs is gone. That means vehicles being in a different place for some players vs other players.
Like you see your car standing on the street, another player sees it inside his own car, his own car suddenly catapults into the sky because "hey there is another car inside me, i need to flobble out of here" even though the car isn't really there
A vehicle, spawned a few cm above ground, floating exactly where it was spawned instead of dropping down the 10-20cm is not a critical vehicle desync making cars explode or bug through walls or make backflips while actually standing still
I actually don't even have the time to help troubleshoot or coordinate feedback collection. I'm just doing it because I want to help and noone else is available.
But all I'm getting is my time wasted by things being reported broken, by people who don't even have the fix installed that we are testing, or the "problem" being reported as not fixed, because a different long not as severe or maybe even unrelated different problem appeared that might not even be connected to the fix we are testing
you need a dedicated test environment reduced to most minimal case
running some complex game mode with random players is certainly not it
I've been running dev branch for over 2 weeks, with all the new things enabled and never encountered problems.
No problem reports in RC, no on dev branch, all fine.
We release to public, dozen life servers report critical problems all of the sudden, we can reproduce one issue internally. We fix it, retest internally, all good. Push fix. Reports come back still just as broken as before.
Seems like the problems only appear in complex game modes. I assume related to server load, networking bandwidth, latency
We had the floating on our basic test also this weekend. Was just a terrain, a few vehicles and few players, no AI and a ZEUS. So would seem to me complex game mode is not necessary, yet I havent followed it in detail. We can try to replicate coming Sunday if thats not too late.
Just checking, but was -limitFPS taken out of the 1.98.146365 new PROFILING branch with PERFORMANCE binaries, v03? I ask only because v1 & 2 were working, but after v3 the FPS is back to being locked on the servers I'm running.
the commandline param works fine with 365 @livid nymph both profiling and performance binaries
Okay, thank you. With most of these things it's user error so I'll triple-check my work. 😀
Just to confirm the assumptions/avoid wasting dev time with frivolous reports, vehicles sometimes floating on ~~spawn/setPos/~~detach did happen before 1.98, we very specifically nudge vehicles after actions like that so they don't float and have done so for a while.
spawn/setPos normally dropped after a small delay. Spawning does however change locality to clients in our mission once vehicle is setup
I do have one question for the devs though, do the syncing issues originate from the clients where vehicles are local, something with the remote client not handling/receiving stuff properly, both or not currently known.
Would help us to determine if the profiling branch builds are having an effect or not
Thank you @thin rain, that info is gold
@thin rain both actually. When local and remote vehicles get close together
isnt the "new" issue a rather big floating offset (aka in meters)?
Yes. And also not just standing still after spawn.
Its driving around, floating away in general around all axis not just standing still in the air
1.98.146368 new PROFILING branch with PERFORMANCE binaries, v04, server and client, windows 32/64-bit, included linux 32-bit server
Fixed: Further improvement of vehicle orientation / position synchronization in multiplayer
anyone else having troubles with diag_captureSlowFrame not working on v04?
executed:
diag_log "frame dump";
diag_captureSlowFrame ['total', 1/300];
the log is in rpt so it evenitively executed but not txt with frame data. it should execute immediately as server had ~90fps and 1/300 should result in capture firing if lower than 300fps
there were problems on server
capture to file maybe worked.
not sure if it always prints the filename in rpt
hmm, it usually worked for me, strange.
didn't change anything on it yet 👀
Am I being dumb, where can I find tbb4malloc_bi.dll?
Arma 3/Dlls
So if I want to try run the performance branch on a dedicated server, grab the 1.98 v4 from the drive, replace the x64 arma3server.exe and then what?
then launch it
Anything I need to change settings wise in order to get the FPS gains?
anything over 200 is bad?
Yeah I'm expierencing that sometimes now, that's why we are looking into this
I mean if I set the limit to say 300 will that cause issues?
probably not. We don't know
Thanks, are you working for BI now 😮
Congrats, last off comment message I promise
I added the limitFPS thing so.. 🙃
Back from my testing that we started publicly over a year ago
Yeah A3FL is struggling big time, we get to 85 ~ players and the server legit dies
well increasing fps limit won't help if you don't even get past 50
is there any way to make the actual process use more computing power, it's sitting at relatively low utilization compared to what we have available?
if there was then we would've already done it
Let us pray for arma 4
there are some undocumented network parameters, maybe they could help if you have desyncs and such. But if your server pulls below 20 fps then that won't help either
We don't really have a lot of desync until the remote calls just stop getting processed and people can't load in
We have been tweaking the config for a while now
server "freezes" but players don't get disconnected?
just see everyone running in place, and can't interact with anything?
they can interact with stuff, drive around and all that sync's - its anything that's being remoteExec that does not work
my exact thought, the server FPS also takes a large hit but people can still move around fine and stuff
Run diag_activeScripts on both server and client what does it show?
Also print JIP queue https://community.bistudio.com/wiki/exportJIPMessages
Server > [3,0,0,0]
Client > [7,0,0,1]
JIP queue is quite long even with one player connected, in the A3PL framework all of the functions are publicVariabled on the client. Is there anything in particular I should be looking at in this file to help find issues?
all of the functions are publicVariabled on the client.
What does this mean?
Not using CfgFunctions in mission.
They define them serverside, and then send all the code to client via publicVariable.
basically to prevent people stealing it, people still did but that's another conversation. I'd say the amount of functions and length of them would cause issues for JIP
We've seen major improvements with the newest performance branch, even from running it only serverside. (~100 Players, 100+ vehicles)
improvements regarding the vehicle desync? or what improvements specifically
Yeah, regarding vehicle desync
major improvements means no problems left, or just very few?
Very few, may it be that this is heavily related to who creates the vehicle ?
if the creator/driver of the vehicle is not on performance branch, he might still cause issues yes
Seems so yeah, we're trying to get more people onto the performance branch, It's just hard to get everyone back onto master afterwards..
While testing we have also found that gunners seem to have no issue but a Co-Driver will massively increase desync.
Which would make sense to an extend. I suppose Co-Drivers are handled different from gunners or passengers since they might be able to steer in choppers atleast.
It's just hard to get everyone back onto master afterwards..
They could just stay on performance branch :3
True, yeah we'll get more people to switch to see if that helps additionally. We spawn most vehicles on the server, should the locality change to the driver be an issue aswell ?
On my server desync seems ok for every one using perf client binary. Still issues with the one using original binary . it seems to happen when we got a mix of players using same car and the driver is not using perf bin
Gonna have to do more testing around that
But no more catapulting cars
and no more vehicle explosion ^^
We're currently adding stats to see how many players are on which branch
should the locality change to the driver be an issue aswell
the change itself no. But the vehicle is simulated where its local. And the issue appears when local
lots of good feedback about using this already
FPS has been as issue for us for years
It's supposed to only improve server fps :u
I've had the same reports from players, ie - higher fps for them.
Don't even try to improve perf. Accidentally noticably improve perf. (╯°□°)╯︵ ┻━┻
Server is also sitting at 100 FPS
Windows server?
yep, about 25 online atm
k
lots of good feedback about using this already
@sterile jewel using what exactly?
I think I have missed something
We've seen no FPS changes neither client or server, atleast not at full cap
Using the performance binary for the server, remember to add the parameter to the server start up
What's the point in running more than 50FPS on the server if it dips far lower when the server fills up ?
What we've seen is that having highest possible FPS result in more time for it to decrease to usual values
Interesting, we usually fall below ~25 at 100 Players after about half an hour
Well, we have less players than this (not life) and also as few additionnal scipts as possible. We use as much stuff as possible the game already provides.
Also scripts used are as light as possible (well thought)
You know, you can write different scripts that result in same function, but one will be heavy and used more times than needed znd the other will be very light and used only when really needed
Scripting is also pretty minimal on our serverside. I think it hardly makes any difference to a milsim server tbh
I think that how often and how many scripts are called also makes a difference
Sure but the fps still deteriorate over time after the server is filled up, without increase in scripting load
We have about 20 server side loops
Sure but the fps still deteriorate over time after the server is filled up, without increase in scripting load
@tame sinew yes, but not as quickly if higher FPS on start
Don't ask me why
I'll try it out and provide feedback
all makes no sense 
We've used a modified .exe for a long time that removes FPS limit completely
not completely if you talking about mine. It just ups to 9k
Not yours
Sure as hell doesn't make much sense yeah
Our main admin modified the exe
But our Server FPS aren't really an issue anyways. We never really dip below 20 and there is no real reason to complain (atleast ignoring the recent desync thing)
Max I've seen was 200+ or 300+ after restart and just with few players connected
With 30 players and hundreds of AI, armor and air, it dropped to 80-90
Our server is 30 slots PvE
Use to restart server every 3 hours on antistasi to get back proper server side FPS. Now not anymore ^^ or just to try the next perf binary and update mods
Server perf still degrades over time, but now it takes longer than before.
And I'm not even talking about latest perf. It was significantly improved some time ago, when 3 FPS bug was mostly fixed
@inland dew
warlords with active mission but 0 players (persistent at #00 server)
Server load: FPS 592, memory used: 1191 MB, out: 0 Kbps, in: 0 Kbps, NG:0, G:0, BE-NG:0, BE-G:0, Players: 0 (L:0, R:0, B:0, G:0, D:0)
warlords with active mission and players
Server load: FPS 84, memory used: 4022 MB, out: 6441 Kbps, in: 687 Kbps, NG:0, G:621, BE-NG:0, BE-G:0, Players: 45 (L:0, R:2, B:0, G:42, D:1)
zeus with no active mission
Server load: FPS 500, memory used: 512 MB, out: 0 Kbps, in: 0 Kbps, NG:0, G:0, BE-NG:0, BE-G:0, Players: 0 (L:0, R:0, B:0, G:0, D:0)
generally any active official server oscilates between 50 and 100 fps with the limitFPS set over 50 , with active playerbase
but that will differ for every server based on hardware and mission type and player load
well, not less than 50 is during how many hours
this was Redux server, where mission just ended (after i took that line not before) , so i would say several hours
we restart our 30 players PvE every 6 hrs for it keep same FPS
don't do timed restarts, only after N mission cycles
but it depends how many players and if the number of them clibs over time or 30 rejoin directly after restart
why?
our mission can take 1 or 2 days to finish it
memory size is no issue on those servers, any leaks or blocks exhaust are bugs which can't be fixed by uploading more RAM 🤣
for the persistent servers i use auto-restart after each endmission or each odd, it works enough well
for the non persisting the value is higher (it's historically agressive but i'm sure even less often restart would be fine)
when we restart at 05:00 p.m. , it's enough to keep good FPS until 05:00 a.m. restart, since past 11:00 p.m., players number starts to decrease, so even if there is less FPS, it goes slightly up when some players leave
since resolving the largest leaks i don't observe visible performance degraded by just run time
runtime, sure, no
but it's number of players and actions going on during a give time
to capture 1 village it can take several hours, some villages take 3-4 hrs if not many players and if they don't cooperate much
that's why i got some of warlords in persist mode to see the degrade
we would have been happy if we could restart only 1x every 24 hrs, technical restart if mission or server config need to be updated
See a life server is a bit different, we don't have any simulated AI or such. Just lots of vehicles.
We manually restart once a day, looking into an automated solution but need to try register a custom command to trigger annoucements
you have already what you need as part of RCON
can you trigger scripted functions from that?
ye well that too, you can use serverCommand
yes, our server is persistent as well and everything is saved, like money, points, number an list of villages already captured, if radio destroyed or not and if not, its placement before restart, to place it in the same place after restart
everything is automated
restarts and what mission to launch/continue, despite restart
i was in this moment indicating default non modded server, ofcourse with callExtension and database you can go into saving everything as it was before
We used a serverCommand to automate restarts, someone dumped our framework using a hacked client and banned every player from the server
so we had to go back to the drawing board on that one :/
even if the radio was destroyed and how much infantry and vehicles and air left and where, so after restart you have to finish cleaning the village, even if no need to look for radio anymore or if just few AI were left there before restart, after restart it will change to next/another village
FPS = tickrate, no?
higher FPS, more frequently the position is updated and the better you can hit the target/enemy?
positions are interpolated even if fewer updates
updates are sent based on error, not based on fps
only when difference of last send, it big enough to current
plus a time factor, but not fps
pretty sure its MinErrorToSendNear that causes updates
yeah for example
smaller value = more updates?
ye
yes
MinErrorToSendNear is for stuff closer to you and MinErrorToSend is for stuff farther away
if anybody is interested
https://forums.bohemia.net/forums/topic/147591-tutorial-server-bandwidth-optimisation/?tab=comments#comment-2605490
https://forums.bohemia.net/forums/topic/147601-all-about-minerrortosendnear/
All about MinErrorToSendNear This thread is part of a much wider discussion on server bandwidth optimisation Tutorial: Server Bandwidth & Optimisation Useful Links BIS WIKI: basic.cfg Read from this post to the end of the thread What the BIKI states MinErrorToSendNear=limit; M...
haven't seen vehicles spawned by server floating in the air even once today, me and server on perf v4
We have about 1/3 of players on perf branch now, we'll continue monitoring for improvements
@sterile jewel what is giving you such a hard time restarting your server ? with a properly setup database just killing the servers process works perfectly fine .. but this is probably not a discussion for this channel, DM me if you have any questions 🙂
Player using new client binary get better stat. One guy just jumped in with the normal binary and blown 2 cars instantly ^^
12:10:38 Server load: FPS 57, memory used: 1852 MB, out: 7687 Kbps, in: 909 Kbps, NG:0, G:231, BE-NG:0, BE-G:0, Players: 63 (L:1, R:0, B:0, G:60, D:2)
15:35:01 Server load: FPS 42, memory used: 2068 MB, out: 7376 Kbps, in: 1051 Kbps, NG:0, G:44948, BE-NG:0, BE-G:0, Players: 72 (L:0, R:0, B:0, G:69, D:3)15:35:21 Server load: FPS 45, memory used: 2068 MB, out: 7255 Kbps, in: 1019 Kbps, NG:0, G:4012, BE-NG:0, BE-G:0, Players: 72 (L:0, R:0, B:0, G:69, D:3) BIG improvement in server FPS
On my server, the starting FPS is over 50, and stabilize at 45 FPS after some hours. It will finally drop around 22 - 23 after a full day of Antistasi Mayhem
that's why we auto-restart our server every 6 hrs, so server FPS and FPS of 2x HCs never drops below 40-50
since everything is saved and restart takes less than 1 minute, players have only to reconnect and continue where they left
- on the same occasion, they're advised to restart Arma on their end, to have better FPS as well, since RAM, vRAM and paging file will also be flushed
helps for some time players, that have only like 8 GB RAM and 2-3-4-6 GB vRAM
Does the performance binary is going to be release as normal binary? Everyone using it are having no desync and if someone with normal binary get on, he feel the desync power 10 000 . One player just blown up all our helis and vehicle in antistasi doing desync ^^
performance/profiling branch is just a testing branch, same as dev branch
everything there will eventually move into release, unless its found to not be stable enough
the vehicle desync is already in RC for the hotfix probably this week
so I did start the profiling branch today, and now memory usage on the server is rising by 1MB/s any advise, except have enough ram? 😄

Ah I think I am just imagining things
@brave granite 1.98.146368 ?
yes
can you replicate it , aka simple mission, nearly no mods or no mods, no players?
it has stopped at 4094 and is stable, I may have not had the proper load on it yet, keeping an eye on it and created data collector sets to see if its leaking anything
32 or 64 bit binary ?
64 bit
So is -limitfps= now part of stable? Or still only in prof?
True, will it cause problems with the parameter when on the stable branch?
Eg. will it cause problems if the parameter is there while not on the prof branch. But it seems to work.
unknown parameters are ignored

