Was this on checkpoint and when you respawned did you respawn from a round-save respawn?

Have you captured C and spawn normally after?

Not sure how we could improve map vote.. maybe trying to implement a plugin but then you got game modes and not sure if thats supported by other plugins.

I'll add, that I've seen the tank spawn bug while trying to capture C. Also, when trying to capture point I everyone spawned inside a truck outside the I bunker. I think on both occasions it was a round end save spawn where the entire team spawned together at once inside the vehicles.

引用自 Josh - Twitch.tv/SgtKillsington：

I'll add, that I've seen the tank spawn bug while trying to capture C. Also, when trying to capture point I everyone spawned inside a truck outside the I bunker. I think on both occasions it was a round end save spawn where the entire team spawned together at once inside the vehicles.

Thats odd. If its happening on round end saves, its because the players position who captured the point was near a tank/vehicle (maybe prone under it) and tahts that position used to spawn the team when everyone dies..

Daiymo you may be right about that. We had gotten B and some were pushing to C. I suspect those in the back may have been at the tank and sniping from there when they died. Will have to see if it happens again. Not sure why the 1 o2 players that were left alive after we typed kill in the console had such bad lag though??

引用自 sheppy：

Not sure why the 1 o2 players that were left alive after we typed kill in the console had such bad lag though??

I took some screens for your analysis, coming from physics heavy source games like garry's mod the teleporting lag without any network choke is server cpu choke. The main symptom of this is "IN" row of the netgraph in the first secreenshot the tickrate drops from 64/s to around 8/s. This helps reduce physics calc, but even then 17 ragdolls hitting each other is too much.

I'm not sure if this is made worse with inertial ragdolls if it uses lua script for physics on top of source engine native physics. But I guess Diamyo can put this to the test by disabling inertial rags and have a full server suicide and spawn under a tank and mass suicide again =p


I don't mod for source so I don't know if native physics in it also uses scripts, so take my guess with salt. You will notice this lag also when we camp bot spawns in Bridge maps, the bots spawn and die so fast on top of each other we create physics hell for ourselves by spawn camping.

https://psteamcommunity.yuanyoumao.com/sharedfiles/filedetails/?id=879396432
https://psteamcommunity.yuanyoumao.com/sharedfiles/filedetails/?id=879395632


Also in the first shot you can see network choke at the beginning, 17 ragdolls flopping causes a network spike and isp throttling but that quickly settles to show only server cpu choke...... perhaps the red colors of the bar graphs can tell us what is eating the cpu?

引用自 🎄Hodor🎄：

引用自 sheppy：

Not sure why the 1 o2 players that were left alive after we typed kill in the console had such bad lag though??

I took some screens for your analysis, coming from physics heavy source games like garry's mod the teleporting lag without any network choke is server cpu choke. The main symptom of this is "IN" row of the netgraph in the first secreenshot the tickrate drops from 64/s to around 8/s. This helps reduce physics calc, but even then 17 ragdolls hitting each other is too much.

I'm not sure if this is made worse with inertial ragdolls if it uses lua script for physics on top of source engine native physics. But I guess Diamyo can put this to the test by disabling inertial rags and have a full server suicide and spawn under a tank and mass suicide again =p


I don't mod for source so I don't know if native physics in it also uses scripts, so take my guess with salt. You will notice this lag also when we camp bot spawns in Bridge maps, the bots spawn and die so fast on top of each other we create physics hell for ourselves by spawn camping.

https://psteamcommunity.yuanyoumao.com/sharedfiles/filedetails/?id=879396432
https://psteamcommunity.yuanyoumao.com/sharedfiles/filedetails/?id=879395632


Also in the first shot you can see network choke at the beginning, 17 ragdolls flopping causes a network spike and isp throttling but that quickly settles to show only server cpu choke...... perhaps the red colors of the bar graphs can tell us what is eating the cpu?

Thanks for posting

What server is that from? We can see if it is the Intertia ragdolls by possibly putting them back to vanilla.

引用自 Daimyo：

Thanks for posting

What server is that from? We can see if it is the Intertia ragdolls by possibly putting them back to vanilla.

Server #1


I have another theory for the lag on those long bridge maps (we had lag on those before you introduced call of duty models).

Travelling salesman algo is computationally expensive no matter which algo we choose, and it should be universally used among bot pathing devs. I've seen bridge maps lag even when bots don't die on top of each other (but still in a spawn camp scenerio).

I suspect this is the second cause of server cpu choke (on certain maps) besides ragdolls. Bots under spawn camp need pathing recalculated constantly, if the map has many pathing nodes the salesman algo becomes bogged down because of the exponential calculation nature of it that goes worse with more nodes. (like when the bots spawn far and there r many nodes between them and the player).