2000fps #sourcekernel

[BehaartesEtwas]

--> http://forums.srcds.com/viewtopic/9126

[DiSTANT]

We have learned to stabilize the kernel quite a bit, but still no 2000 fps kernels yet.

[BehaartesEtwas]

I still dont see what you want to gain from 2000 fps. The server will run worse that with 1000 fps on the same hardware, that's almost for sure. You will lower the theoretical ping limit by 0.5 ms, but you probably will never achieve the same stability as with 1000 fps. I think you will come close to the precision limit of the clocks in most PCs, it would need to be below a microsecond (1us are 4 fps at 2000 fps). I don't even want to think about all the wake-up latencies etc. in kernel process handling. You basically double the effekt of all this. That is, in case you really manage to breake the 1000 fps limit in the engine, without introducing bugs like the speedups we know from hlds with pb3 and sys_ticrate 10000.

[dualcore1289]

the theory about the fps, is that it will run the game smoother, better reg, etc. But there is a limit to it. Again as mentioned, theres no real difference between 1000fps and 2000fps besides 1000fps difference. 2000fps speeds up game play however. And the human eye cant even see past 400-500fps anyways. And to everyone who says they notice a difference, that may be so on a crappy server, overloaded server boxes, or just plain rates are not set right. It is nothing but a sales tactic.

[BehaartesEtwas]

dualcore1289 Wrote:And the human eye cant even see past 400-500fps anyways.

the humal eye can't see any difference past 40-50 fps (one "0" less!), but thats a different story (btw: movies have 24 fps).

the difference is, that a net engine needs to compensate for any latencies. that has serveral implications, in particular you need mechanisms like an interpolation of the data. In theory, the more often you calculate the more precise the prediction (you don't need to look too far into future). Also, in theory your predictions are more predictable ;-) if you do them regularily (i.e. your fps do not vary).

I bet nobody will notice any difference between 500 and 1000 fps, provided that both are running very stable. I doubt anyone has done any reliable test (you would need to do a blind testing, i.e. not give the test subjects the rcon pw, and e.g. switch between 500 and 1000 fps randomly after each round). But what I definitivly know is that stability has a big impact. A server with fps varying between 500 and 1000 fps is definitivly worse than a server running with constantly 500 fps.

[Saturnsix]

i cant wait to get a 4000fps stable so i can brag how mine is bigger than yours.


what has the world come to

[get]

pls, on having near 20 players gaming on your server do a stats, and also do on the shell a "uptime" to see how much this server takes from the system and if its good idea to have it so.

[css]

Do the math:

1000 FPS slot = $1
4000 FPS slot = $5

Do 20*$5 on calculator to see how much this takes from customers and if its good idea to have it so.

[BrutalGoerge]

i did a blind test and varied the fps from 100 for about an hour, then to 999 for an hour.

and GHASP

No one could tell anything was going on.

I was kind of disappointed, all that tweaking and mucking around with settings for nothing.

[BehaartesEtwas]

BrutalGoerge Wrote:No one could tell anything was going on.

What kind of players? I think it makes a difference if you try this on a public server or with some pro players during a war...

[iliketohostservers]

http://people.summit-servers.com/monk/archives/81

Monk, the owner of Summit Servers, and a master Unix GURU claims that theres only 2 ways to achieve 2000fps... basically LIE to the engine (trick the game engine) and setup a proper 2000hz system which isnt easy to do...

He says that 2000hz systems have extensive problems like clock skewwing and drifting. So basically the only working way to get 2000fps is to trick the game engine...which is pointless because it wont make your gameserver any bertter. pz.

[BehaartesEtwas]

Well, he is wrong:
http://fpsmeter.fragaholics.de/?load=srcds.127.0.0.1.20015.1232214721.session

I did not lie to the engine, it is really running at 4000 fps (but of course I need to trick it to run faster than 1000 fps ^^). And I did not need to patch the kernel HZ, the server is even running with only HZ=100 and dynamic ticks...

Nevertheless in one point he is right: it is completely pointless. The server will probably run worse than a good and stable 500 fps server.

[iliketohostservers]

BehaartesEtwas Wrote:Well, he is wrong:
http://fpsmeter.fragaholics.de/?load=srcds.127.0.0.1.20015.1232214721.session

I did not lie to the engine, it is really running at 4000 fps (but of course I need to trick it to run faster than 1000 fps ^^). And I did not need to patch the kernel HZ, the server is even running with only HZ=100 and dynamic ticks...

Nevertheless in one point he is right: it is completely pointless. The server will probably run worse than a good and stable 500 fps server.

If the hz is only 100 then you had to trick the game engine in some way, therefor it isnt a true 4000fps. The hz limits the FPS unless you trick the game engine.

[BehaartesEtwas]

iliketohostservers Wrote:If the hz is only 100 then you had to trick the game engine in some way, therefor it isnt a true 4000fps. The hz limits the FPS unless you trick the game engine.

No thats not right. hrtimers are working different.

[iliketohostservers]

BehaartesEtwas Wrote:

iliketohostservers Wrote:If the hz is only 100 then you had to trick the game engine in some way, therefor it isnt a true 4000fps. The hz limits the FPS unless you trick the game engine.

No thats not right. hrtimers are working different.

cereal? didnt know that , sorry lol