1000 FPS + on Debian 5 please

[Monk]

Don't expect everything to behave the way you want it to with sourcetv enabled. When you have an app write to disk, it has to constantly open()/read() the file it's writing to, causing huge latency, which makes.. your server's FPS bounce around. Just a design issue, there's not a dedicated thread for sourcetv, like the old 1.6 stuff.

[WolwX]

(11-18-2010, 08:32 AM)Monk Wrote:  Don't expect everything to behave the way you want it to with sourcetv enabled. When you have an app write to disk, it has to constantly open()/read() the file it's writing to, causing huge latency, which makes.. your server's FPS bounce around. Just a design issue, there's not a dedicated thread for sourcetv, like the old 1.6 stuff.

So that mean people having perfect result while players on the server are doing that without STV On ? :s

Anyone can confirm ?

I will test without STV on my side ...

[lol554]

I can confirm this.

[BehaartesEtwas]

yes, sourcetv isn't very well implemented unfortunately... I know of no publicly know solution for this problem.

[Terrorkarotte]

Things to reduce the impact:
- Use a fast datasystem and hdd (ext2 is not the fastest )
- Start STV with onl 1 slot and connect a proxie on that slot. The Proxie will do the Broadcasting

[WolwX]

Ok

So here are my test :

Stock Kernel Debian + idler + resched.sh


and one test with Kernel 1000HZ RT + idler + resched.sh


As you can see, after the map change, the full fps drop, there some little spike cause because I readded the STV on the server ...

F**k this TV ><"

That suck because the recording of STV demo is really important ...

So finally I will try to find the better settings for my STV ...

[Terrorkarotte]

You do not want to understand what we said over and over agian right?

HIGH FPS ARE USELESS AND A WASTE OF CPU POWER.

But hey how cares if a person just wants to make money out of it and sell things people do not need.

If this person keeps on telling his customers you need more fps and pay more even if he knows better it is kind of a fraud.

[WolwX]

(11-20-2010, 08:08 PM)Terrorkarotte Wrote:  You do not want to understand what we said over and over agian right?

HIGH FPS ARE USELESS AND A WASTE OF CPU POWER.

But hey how cares if a person just wants to make money out of it and sell things people do not need.

If this person keeps on telling his customers you need more fps and pay more even if he knows better it is kind of a fraud.

And you still want to be like that with me ?

I want to reach a 1000 FPS stable then after I will take a final decision about my configuration, that's the subject of my topic.

Your precedent post was util, this one still useless.

Respect my choise and my question, I already understood your opinion but that's not the subject of this topic.

[WolwX]

For people looking how understand the high FPS settings :

https://support.steampowered.com/kb_article.php?ref=5386-HMJI-5162 Wrote:WHY HIGHER FPS?

The key reason to run higher FPS is the render time. At 1000FPS, the server is rendering one frame every 1 millisecond (ms). This means that the worst-case adder to the player ping is only 1ms, IE: the player gets more accurate data and can get it more often.
At 300FPS it's only 3ms which is perfectly acceptable, but at 100FPS it's 10ms, which is a significant percentage of a 100 ping (10%). A player with a 100 ping would actually be getting 110ms response time from the server. Many AMD systems will only run 60FPS without the ping booster which is 17ms.
This is still not too significant but it can change the feel and response time of your server for players.
Without the FPS Boost your server will use significantly less CPU but accuracy may suffer.

So I opened another topic for that => http://forums.srcds.com/viewtopic/15683

That's simply for information

[Monk]

I think I agree with terror, you're wasting time trying to do all these tricks to get 1000FPS stable. While I respect your choice to run it because you think it's better, I think it's a worthless attempt to justify the time spent on trying to get the results you want when you;

- Have OS latency to contend with
- gettimeofday latency / not counting at a fixed rate (even with TSC or HPET driving it)
- Have a closed source engine filled with issue (ie: sourcetv, writing to a directory with huge amounts of files in it)


FWIW there is no way at all you/I could tell the difference in a clinical trial.

The article you 'posted' doesn't mention anything that was already known, it's based on an old article that was published by a non valve-employee and added to their support section. It's not an official response from valve on it. Source is different than OB.

[lol554]

(11-21-2010, 01:31 AM)WolwX Wrote:  For people looking how understand the high FPS settings :

https://support.steampowered.com/kb_article.php?ref=5386-HMJI-5162 Wrote:WHY HIGHER FPS?

The key reason to run higher FPS is the render time. At 1000FPS, the server is rendering one frame every 1 millisecond (ms). This means that the worst-case adder to the player ping is only 1ms, IE: the player gets more accurate data and can get it more often.
At 300FPS it's only 3ms which is perfectly acceptable, but at 100FPS it's 10ms, which is a significant percentage of a 100 ping (10%). A player with a 100 ping would actually be getting 110ms response time from the server. Many AMD systems will only run 60FPS without the ping booster which is 17ms.
This is still not too significant but it can change the feel and response time of your server for players.
Without the FPS Boost your server will use significantly less CPU but accuracy may suffer.

So I opened another topic for that => http://forums.srcds.com/viewtopic/15683

That's simply for information

That information is outdated. It were written a long time before CS:S were converted from the HL2 engine to the OB engine, so in fact, this doesn't say anything at all now... Don't waste your time on 1000 FPS.. seriously..

[saintjimmygd]

Seems it's true (tickrate 66, fps_max 100, host_framerate 0, no source-tv).Also,kernel-2.6.33-zen3 with BFS,100 Hz Preemptible,Server envoirement:



In-game,my ping is 5 (the same as when i ran 1000 fps) and game-play is smooth.
What about HLDS servers ? Is there any difference between sys_ticrate 1000 and sys_ticrate 100 ? Because when i run my HLDS server with sys_ticrate 100 (and pb3) my ping is higher then with sys_ticrate 1000 (and pb3).

[Monk]

Probably because pingboost 3 uses a different method completely to do timers, it's an alarm on each packet that comes in. So you get different ping results depending on the HZ of sys_ticrate...

Of course, if ping is influenced by this much stuff, I wouldn't trust it anyways, lol. Layer7 latency calculation -always- sucks. If the server is busy processing frames, it wont have time to update the ping, causing higher pings...

[BehaartesEtwas]

it's more vice versa. if the server is busy processing frames (or sleeping until the next frame in case of hlds and pre-orangebox srcds), it cannot receive network packets, thus actually increasing the overall latency, but it still believes the latency is constant (as it cannot derive the latency of a single game packet).

still the ping in the score board or net graph has little to do with the actual latency of the game packets. only the rcon status command shows the real latency, at least on srcds. actually I don't really know how which ping is measured. apparently there are at least 3 different methods (score board, net graph and rcon status). if anyone has more information...

[Terrorkarotte]

With pingboost 3 it is easy to achive a 1000 line if you have the bfs scheduler: +sys_ticrate 1001 or 1010.
With pingboost 2 or none you will have around 950-980 fps like with srcds. Since pb 3 uses much cpu which the other do not and no player can tell the diffference between the modes I would reccomend against using pingboost 3. The small benefit from 950 to 1000 fps does not justify a cpu usage that can be more than twice as large.