1000 FPS, network usage and others.

[css]

HBS|Ryan Wrote:I'm not going to install your plugin on my commercial servers. I might on my personal server.

Install it on your personal server and have it monitor your commercial servers. All it does is run "stats" command five times in a row every five minutes. The code is open source, so you can inspect the code before you run it.

You may have got offended about me being suspicious. Don't be. There are other GSPs who evidently have advertised their non-1000 fps servers as 1000 fps servers. Being suspicious in this case is just common sense.


The stats output you posted seems valid, but as long as it's just bots without any network traffic, it's just simulation.

Here's some proof to my case too:

Stats from server with 10 bots. There are also two other game servers running on the same server. One server is 24 slot / 66 tick and another is 100 tick empty awp server. Anyway, here are stats for the test 10 bot server:

Code:

stats
CPU   In    Out   Uptime  Users   FPS    Players
38.33  0.00  0.00       0     0  841.75      10
stats
CPU   In    Out   Uptime  Users   FPS    Players
38.33  0.00  0.00       0     0  958.77      10
stats
CPU   In    Out   Uptime  Users   FPS    Players
36.25  0.00  0.00       0     0  941.62      10
stats
CPU   In    Out   Uptime  Users   FPS    Players
36.25  0.00  0.00       0     0  957.85      10
stats
CPU   In    Out   Uptime  Users   FPS    Players
36.25  0.00  0.00       1     0  298.33      10
stats
CPU   In    Out   Uptime  Users   FPS    Players
36.25  0.00  0.00       1     0  265.32      10
stats
CPU   In    Out   Uptime  Users   FPS    Players
36.25  0.00  0.00       1     0  258.40      10
stats
CPU   In    Out   Uptime  Users   FPS    Players
34.40  0.00  0.00       1     0  961.54      10
stats
CPU   In    Out   Uptime  Users   FPS    Players
34.40  0.00  0.00       1     0  890.47      10
stats
CPU   In    Out   Uptime  Users   FPS    Players
34.40  0.00  0.00       1     0  904.16      10
stats
CPU   In    Out   Uptime  Users   FPS    Players
34.40  0.00  0.00       1     0  577.70      10
stats
CPU   In    Out   Uptime  Users   FPS    Players
30.00  0.00  0.00       1     0  242.07      10
stats
CPU   In    Out   Uptime  Users   FPS    Players
30.00  0.00  0.00       1     0  928.51      10
stats
CPU   In    Out   Uptime  Users   FPS    Players
30.00  0.00  0.00       1     0  734.75      10

There are no timestamps, but I had the server running and was doing UPARROW-ENTER-UPARROW-ENTER style of "stats" flooding. The timespan is about 2-3 seconds. You can see that the server hits 900 FPS (with kernel CONFIG_HZ 100) quite often even with 10 bots there.

Because there are the two other servers also running, they might affect the FPS heavily. Here are stats for empty server, so this should show how much the other two servers affect the FPS:

Code:

stats
CPU   In    Out   Uptime  Users   FPS    Players
4.00  0.00  0.00       1     0  882.61       0
stats
CPU   In    Out   Uptime  Users   FPS    Players
4.00  0.00  0.00       1     0  958.77       0
stats
CPU   In    Out   Uptime  Users   FPS    Players
4.00  0.00  0.00       1     0  764.53       0
stats
CPU   In    Out   Uptime  Users   FPS    Players
4.00  0.00  0.00       1     0  813.67       0
stats
CPU   In    Out   Uptime  Users   FPS    Players
4.00  0.00  0.00       1     0  819.00       0
stats
CPU   In    Out   Uptime  Users   FPS    Players
4.00  0.00  0.00       1     0  966.18       0
stats
CPU   In    Out   Uptime  Users   FPS    Players
4.00  0.00  0.00       1     0  961.54       0
stats
CPU   In    Out   Uptime  Users   FPS    Players
4.00  0.00  0.00       1     0  968.99       0
stats
CPU   In    Out   Uptime  Users   FPS    Players
4.00  0.00  0.00       1     0  102.17       0
stats
CPU   In    Out   Uptime  Users   FPS    Players
4.00  0.00  0.00       1     0  955.11       0
stats
CPU   In    Out   Uptime  Users   FPS    Players
4.00  0.00  0.00       1     0  958.77       0
stats
CPU   In    Out   Uptime  Users   FPS    Players
3.75  0.00  0.00       1     0   98.63       0
stats
CPU   In    Out   Uptime  Users   FPS    Players
3.75  0.00  0.00       1     0   22.11       0
stats
CPU   In    Out   Uptime  Users   FPS    Players
3.75  0.00  0.00       1     0  548.25       0
stats
CPU   In    Out   Uptime  Users   FPS    Players
3.75  0.00  0.00       1     0  559.28       0
stats
CPU   In    Out   Uptime  Users   FPS    Players
3.75  0.00  0.00       1     0  963.39       0
stats
CPU   In    Out   Uptime  Users   FPS    Players
5.20  0.00  0.00       1     0  958.77       0
stats
CPU   In    Out   Uptime  Users   FPS    Players
5.20  0.00  0.00       1     0  677.51       0
stats
CPU   In    Out   Uptime  Users   FPS    Players
5.20  0.00  0.00       1     0  959.69       0
stats
CPU   In    Out   Uptime  Users   FPS    Players
5.20  0.00  0.00       1     0  949.67       0

The FPS drops even on a empty server, so clearly the other two servers were affecting my test results. However, I think if I had had only the 10 bot server running, then the FPS might have been stabler around 950. Sadly I don't have resources to test what it would be on a dedicated 10 bot server. I've only got one physical server.

Anyway, with this experimental evidence gathered from my test server, I conclude that the plain "stats" dump for few seconds on a 10 bot test server is not enough to prove any server stable ~1000 fps.

HBS|Ryan Wrote:your more then welcome to message me and get one then when its full run stats until your happy

Thanks, but I'm here on the other side of the sea. Latency would be too high.

Anyway, if you've got test server running for demo purposes, I can set up monitoring on that server. Then we'll see what kind of long run FPS numbers the server really does.

[Spartanfrog]

css Wrote:The stats output you posted seems valid, but as long as it's just bots without any network traffic, it's just simulation.

Actually FPS is not affected by network traffic, only tickrate .

[HBS|Ryan]

Our servers are for competitive play and can not have plugins installed other than zblock. But, I'm not really interested in your statistics add on, so I am not going to install it. but if anyone doubts the sweetness of http://www.HotBettyServers.com, just join our IRC channel and pm me. You can test one out yourself.

[css]

Spartanfrog Wrote:Actually FPS is not affected by network traffic, only tickrate

You say this without any proof. Why should I believe you?

Try referring to Whisper's wiki or some other reliable source of information.

I think you've understood the tickrate's effect to server traffic wrong. You've probably read in Whisper's wiki that higher tickrate requires more bandwidth. Then you think that vice versa higher traffic affects tickrate also. Is this how you came up with your ungrounded claim?

HBS|Ryan Wrote:Our servers are for competitive play and can not have plugins installed other than zblock.

As I said, you can install the monitoring system for example to your home computer. Then set it to monitor your production server. The production server will only see five RCON "stats" commands every five minutes. There is no need to install anything on the production server. The production server will stay "as is", but you'll get FPS statistics from it.

[Spartanfrog]

I've actually referred to Whisper's Wiki plenty of times so I really don't need snide comments from you. I've done my own tests and guess what? Higher FPS doesnt affect network usage. I could care less if you do not believe me. If you want to make snide comments on people who are trying to help you and you are going to ignore obvious information, All I can say is whatever floats your boat.

[Drocona]

I'll list it, what does higher FPS do:
- Increase CPU usage
- Increase smoothness in the server
- Increase (in some small cases) hit registration

FPS has no direct relation to bandwidth, it can be indirect by which I mean:
FPS drops below tickrate, tickrate comes down because of low FPS resulting in less bandwidth transfer due to the tickrate drop.
I've tested with 40 players, no matter if I had 80 FPS (low cpu usage) or 512 FPS (high cpu usage) the bandwidth output was the same, at 33 tick an average of 12.20KB/s per slot, spikes to 16.5KB/s per slot.

Also Whispers page is not law, it only states theoretical calculations and information, also it is pretty outdated. Whispers is nice to get a global view on servers but hardly states any facts at all that have been proven running a server real time. I'm not saying all is wrong, if you don't know what you are doing it will explain you the basics but there's a lot more to it in the non-text world of running servers.

[css]

Spartanfrog Wrote:I've done my own tests and guess what? Higher FPS doesnt affect network usage.

It's not what I said. Let's try to stick to the issue and not talk about three things at the same time.

1. I said: Network traffic might affect FPS
2. You said: Network traffic affects only tickrate
3. I said: Tickrate affects network traffic

See the subtle differences there. Especially the "tickrate -> traffic" and "traffic -> tickrate" difference. The latter conclusion is wrong, although the former holds. "Network -> FPS" is somewhat undecided yet


Then few words to reason why network traffic might affects FPS, but not tickrate.

When the game is run only with bots, the server does not need to concern itself with interpolation - or maybe it does, it's about how the bots are designed. Anyways, I assume that the server simulates bots in the most efficient way so that it doesn't have to do lag compensation or think about how much bots have interpolation. This might have affect on the FPS when comparing to playing with normal internet players.

Whisper says that

Code:

Server FPS
    Because there are no keyboard or mouse I/O's occurring, it only deals with how often the server checks for (incoming) game packets.

See the "Lag compensation" in Valve wiki. It seems to be lot of work to "go back in time" everytime when somebody moves or does something. Also "This moving backwards must account for both connection latency and the interpolation amount the client was using that frame." With bots there is no connection latency or interpolation. Bringing network to the game must have some kind of effect to it.

Tickrate: During each tick, the server processes incoming user commands, runs a physical simulation step, checks the game rules, and updates all object states. After simulating a tick, the server decides if any client needs a world update and takes a snapshot of the current world state if necessary. A higher tickrate increases the simulation precision, but also requires more CPU power and available bandwidth on both server and client. [ http://whisper.ausgamers.com/wiki/index.php/Tickrate, http://developer.valvesoftware.com/wiki/Source_Multiplayer_Networking#Basic_networking ]

It's not reasonable to assume that the server would cut tickrate instead of FPS when the game gets CPU intensive. As long as the FPS is over the tickrate, the server can receive and send enough packets to satifsy all players. The downside in low FPS is probably that the received game packets do not have accurate timestamps. That's because the server does not have time to check for incoming packets as much as it'd want. Thus, it receives bulk of network traffic at the same time and marks them with the same timestamp. Then in the game simulation it suffers from this inaccuracy.


BTW. Whisper's wiki and Valve wiki are the two most reliable source of information in the 'net.

About the 1000 fps thing in general, it's just a marketing gag. (Note: The author says in the last chapter before "Conclusion" that the server would always try to reach the maximum FPS by cost of other resources - ie. tickrate. I can't agree with this, because on my server I've got fps_max 600, which never happens, but yet players can get 66 updates per second. Rest of the stuff about not noticing anything after 50 ms network latency is true )

PS. The easiest way to resolve the issue here is for Ryan (or any other 1000 fps server admin) to show long run (eg. 1 day) stable 1000 fps statistics for popular ~16-24 slot 66/100 tick server.

[HBS|Ryan]

Those articles are old and written before hardware could handle 1000 FPS and greater.

I have lost all interest in proving my claims to you. Thanks for trashing my thread and my company for the sake of being a know it all, when your experience is lacking.

If anyone else is interested, you know how to reach me.

[css]

HBS|Ryan Wrote:Those articles are old and written before hardware could handle 1000 FPS and greater.

Which articles you're referring to? The only one that concerns specifically 1000 fps is m.freaks|BD|Lefty's (moreal.org) article. The rest of the links are describing the Source engine specifications in general. Moreover none of the articles say anything about hardware resources. While reading the articles you can assume that they have super powerful computer from the future.

HBS|Ryan Wrote:I have lost all interest in proving my claims to you. Thanks for trashing my thread and my company for the sake of being a know it all, when your experience is lacking.

I'm not trashing your company.

In my previous message you must see that my experience is not lacking. I've proven my claims.

You've got to be ready to talk about your high quality servers. I asked for proof that your servers can do the claimed 1000 FPS. In my experience all servers lose the 1000 FPS when players join there. Unfortunately you take it as offence. It was not my intention.

Anyways, you clearly have reached 1000 FPS on a empty server. That's great. It's still unclear whether the FPS remains while there are players playing. In a marketing point of view you've got enough grounds to sell your servers as 1000 FPS.

[HBS|Ryan]

FPS isn't effected by any part of network connectivity. Its a purely hardware/configuration based.

bots take up more system resources than real clients connected.

i showed you a server full of bots my stats output. you responded with an empty server stats that bounced all over the place. that should tell you alone that it solid. Other respected members of this community will agree.

Your determination to be right in this case will not help you. I'm sorry.

I originally joined this forum to get help from the experienced. Reading articles that were a year+ old wasn't answering my questions. After about 18 months of testing and playing to find the right configuration, I got what I was trying to get out of the forum. Now i try to help others. Im not really here to sell servers, as my client base wants to buy servers rather than make servers. Thats why im not in every thread posting about HotBetty.

Now it seems your purpose on these forums is not to get help or even help, but to promote your plugin.
Though your plugin/app is a nice one for people who are that interested, some of us aren't.

I know as well as my customers know the quality of the service we provide them. I'm sorry you have an issue with that, but thats the way it its gonna be. I'm done trying to explain our services to you.

if you want one for yourself buy and opteron 2000 series and a nforce motherboard. Ill be happy to compile a kernel for you.

[Spartanfrog]

I agree with Ryan and like I stated before, FPS is not affected by network usage! If you dont know that and refuse to believe us because you believe yourself right....well good luck with that.

[ComputerHelper]

Wow this kid needs to do some up-to-date research, owning my own GSP-in-the-works FSP is no where near affected by network usage, the thing that effects ur FPS the most is your HARDWARE!!! nub

[ClanWoof]

All I can say is when I tried out HotBettyServers. I tried one out for three scrims. I also typed Stats in console a few times because myself I wonder. Everytime it was never lower then 990 FPS. Everything ran like they say.

[css]

ComputerHelper Wrote:Wow this kid needs to do some up-to-date research, owning my own GSP-in-the-works FSP is no where near affected by network usage, the thing that effects ur FPS the most is your HARDWARE!!! nub

Take your time and try to come up with better answer. You refer to "network usage" as if it was some mystical resource that you let come into your server. Players generate network usage.

What I've been saying in all my posts, is that when players join a server, the server FPS usually goes down. None of you have provided any proof that your 1000 FPS servers can keep up the 1000 FPS when there are 16-24 players playing. I've had chance to monitor couple "1000 FPS" servers which turned out to not be 1000 FPS when players played there. I've got my reasons to be suspicious here too.

Here are stats for my server.



You can see that the FPS fluctuates between 500 and 400. Those are one week stats. If I type "stats in console a few times" when the server is not full (ie. ~5-10 players), I'm quite likely to get FPS around 500. Then again, if I type stats when the server is full (24 players), I probably get 400-450. See for example 27th day. The thick green line (avg) is all the time over 450+. Anyways, it's very clear that the FPS is affected when the server is running full 24 players. I believe this happens with 1000 fps servers too, but it's not noticeable when running only couple "stats" command every now and then.

The server FPS in the above graph is also affected by other load on the physical server.

The server is one of the world's TOP-200 biggest CS:S servers according to gametracker.com and game-monitor.com.

If you kids have something to say, say it with proof. Talk the talk, walk the walk.

Note: HotBettyServers.com does have 1000 FPS on empty servers. It's good indication that the servers are very high quality.

[ClanWoof]

"If you kids have something to say, say it with proof. Talk the talk, walk the walk."

First off get your info right. You don't have 24 players on a 10 man scrim server.
Also like he said go to their irc channel get a test server have a few scrims on it with real players, you would get the same proof that I did myself when I tested out their servers with 10 Players in a scrim.
Edit made it more nice.