Server likes to shit itself with relatively high playercounts.







  • I did post about this in the broken stuff thread. It’s a real issue.



  • I spoke to Box regarding this. Not a lot that can be done as ZS is a resource hungry GM.

    I have spoke to JetBoom in the past regarding this. We found changing the nailing system helped a lot. We were running ~50 with reduced lag.

    The only way to remove this issue is to streamline the ZS gm. This is something me and the zs2 team are trying to do at the moment but it isn’t a easy thing.



  • This also happened on zs_gauntlet_reborn_b5, where there was no props but a lot of players >50



  • By the way, is this only with ZS? Does EFT do it with a full server or is the player limit not enough? I thought it might be the mysql checking the status of the connection every 15 seconds but removing it didn’t seem to help, at least according to box. I’m going to install a profiler and see exactly what calls the server is choking on.

    @Swingflip:

    Not a lot that can be done as ZS is a resource hungry GM.

    false.

    I have spoke to JetBoom in the past regarding this.

    false.

    We found changing the nailing system helped a lot. We were running ~50 with reduced lag.

    false. There is nothing to change. The nails disable physics on anything they’re nailed to and are completely event-driven.

    The only way to remove this issue is to streamline the ZS gm.

    I told you already that it’s already as streamlined as it’s going to get. Anything you do is most likely just going to make non-existent problems become a reality.

    zs2

    You need to stop, seriously. I told you to stop before, now you’re just being retarded. I don’t care if you plan to release it or not. It’s not a sequel, it’s not a version 2. Relay this message to the other guy that’s making a “zs2” server and messaged me but never responded to anything I said.



  • @JetBoom:

    Does EFT do it with a full server or is the player limit not enough?

    EFT never has a problem even when it’s full.



  • ehhh, could you please limit the cam2D3D depending of the length of the eye traces? Already tested and arsenal crates/resupply/aegis  is incredibly expensive  in a high amount for the client to render ( it could have something todo with this issue… ) The perfect length would be about… 150- 200 units, after that, don’t draw it.



  • ehhh, could you please limit the cam2D3D depending of the length of the eye traces?

    What traces? 2d3d isn’t even that expensive. EFT relies heavily on them but doesn’t suffer from them.

    @Robbert:

    ( it could have something todo with this issue… )

    It doesn’t. I could force you to draw 1000 of these and the server won’t care.

    The perfect length would be about… 150- 200 units, after that, don’t draw it.

    That’s basically 1.5x a sideways player’s distance away.



  • @Robbert:

    cam2D3D

    incredibly expensive

    heheheheh

    heheh

    heh

    heh

    if you’re talking about how your 9999 fps goes down to 45 when you look at one that’s not the kind of lag we’re talking about.

    We’re talking about server lag. Lag everyone gets regardless of who has better computers.

    You’re talking about client lag. Lag certain people get based on how good their computer is.



  • Silly me, it is indeed client side but you do have fps drops/spikes with allot of them so the request stay’s. ( even though it’s not relevant to the issue… )
    My best guess is that something goes wrong with initializing, other servers with loads of players and other activity’s have this aswell. it’s not just ZS or noxiousnet.

    Also, I did a test with box and raox yesterday to see how many times ShouldCollide calls and we found out that, with about 100 nextbots ( the effect with players are even worse ) it can call about 2 million times. It din’t do this before ( at least not this bad )  so that might explain some lagg.



  • 2 million calls per what?

    ShouldCollide is expensive, sure, but it’s required for a lot of things and would cause overall degraded performance if that was the culprit. It would not cause the server to choke every X amount of seconds - that’s probably something running on a timer that’s O(n^x) depending on the amount of players.

    I have the profiler ready to go - if someone reminds me when the server has 50 or so people I can check. Otherwise I’ll wait until prime time tonight.



  • Box pointed something out called dbugger, it’s not perfect and doesn’t really give good information but it did show me the calls/s of the function. ( something todo with slowing it down? ) anyway, it showed me the immense amount of calls on that function, this was on sandbox. And it The server almost crashed at that moment. But i’ll hope you find the issue! Mind sharing if you do? I am curious what is causing it.



  • So you profiled sandbox and found out it called this function a lot, and thus ZS calls the function a lot too?



  • i would assume it does since sandbox doesn’t do anything with it, it just returns true and that’s it. Already asked around and it isn’t supposed to call this many times therefor something went wrong on the last update ( or before that? )
    so, if allot of humans are on zs in the same spot i guess that’s something causing allot of lagg. Also, do you still use CollisionRulesChanged() stuff? if so, that might be something too. Just got told it never was there, i read something on FP about you using it, sorry for the inconvenience.



  • Box pointed this out to me: https://github.com/Facepunch/garrysmod-issues/issues/1168

    If it is ShouldCollide, then it’s beyond fixing for until the next GMod patch. I have no idea how long this has gone unfixed for.



  • When fixing a problem, it’s probably best if we don’t throw guesses towards the wall to see what sticks, let’s see what the profiler says.



  • @Raox:

    Box pointed this out to me: https://github.com/Facepunch/garrysmod-issues/issues/1168

    If it is ShouldCollide, then it’s beyond fixing for until the next GMod patch. I have no idea how long this has gone unfixed for.

    There’s no cases where I turn it off though. I always thought it was set on and forget it.



  • Gauntlet reborn is laggy as shit because of the map. Lighting is buggered also it tries to force models and textures upon the client which bugs out the FPS so that is a separate issue.

    Sql won’t cause lag spikes at high player counts. However MySqloo is a pain in the arse and does cause issues depending on how it’s used. Usually this is the case when player’s connect and disconnect.

    If the Sql script is written correct it won’t lag out when updating. Our SQL updates every 5 seconds with no noticeable lag.

    The prop issue does cause problems but is probably not the issue in this matter. Do you find the spike is more noticeable on player join/dc? We just tried our ver of zs and were running 45 players with no issues on Mall.

    More to the point. What maps is this spike occurring on? Also what cl errors are you receiving on map load again it could be a case like gauntlet reborn and bad texture/model issues.

    Out of the 50 most common zs maps out there I have only really found around 10 which don’t cause issues at or around 40-50 players.



  • Did you not read anything at all?


Log in to reply
 

7
Online

11.3k
Users

15.5k
Topics

300.1k
Posts

Looks like your connection to NoXiousNet was lost, please wait while we try to reconnect.