Help me Find a good Server Config for my Specs

[GossamerSolid]

Right now we have problems that we didn't seem to have problems on our old server (same specs, but we had different configs back then. Long story short, we no longer have possession of that server so we lost the configs).

We're looking to get a config that works better for our server (less strain on the CPU mainly). We're definetly getting more server FPS issues than before.

Server Specs:
Xeon 4 Cores @<hidden> 2.27Ghz
4GB DDR3
100mbit Connection

Here's my current server config:



MinBandwidth=15000000;
MaxBandwidth=100000000;
MaxMsgSend=1024;
MaxSizeGuaranteed=1024;
MaxSizeNonguaranteed=64;
MinErrorToSend=0.0024999999;
MaxCustomFileSize=0;



Can anybody chime in with some improvements we could make to our server config?

NOTE: These are the values from the Kelly's Heroes A2 server tutorial.

[VisceralSyn]

Well...

Since you have posted the pipe is 100Mbps, why would you use 15Mbps as a minbandwidth? Even guessing for overhead should be between 88 and 92Mbps.

Bandwidth is typically the issue, you should have little to no problems.

I've been doing some research on the MaxMsgSend, and the MaxSizeGuaranteed, MaxSizeNonguaranteed. If MaxMsgSend is actually how many packets the server can send, in one simulation cycle, and MaxSizeGuaranteed is the packet. Then MaxMsgSend can be a very high number.

Example 100Mbps can send 12,207 1,024byte packets.

Somewhere I read that smaller packets is better for gaming than larger packets, and that larger packets are better for file transferring. But, thats using TCP, UDP is mostly used for gaming, and is connectionless and does not ack. With this, I have been experimenting with:
MaxMsgSend=2048;
MaxSizeGuaranteed=128;
MaxSizeNonguaranteed=64;

Because my upload speed is 2.2Mbps, which should be capable of 2148 packets at 128byte size. Giving myself a little headroom, there.

Then I also read somewhere, that TCP treats packets smaller than 64bytes, as a fragmented packet. Which would be bad for gaming, but again we're using UDP here. So, with that in mind, MaxSizeNonguaranteed shouldn't be less than 64bytes.

Hopefully someone else has more information, because mostly, the config file lets you fine tune network settings, and not really how the cpu runs the server. Pretty much go with the fastest CPU you can get and goo-gob's of ram!

...Syn...

[GossamerSolid]

I'll try that MaxMsgSend value you posted.

But you said that with high bandwidth you should send larger packets, why are you using 128 and 64 for those MaxSizex values?

[VisceralSyn]

Because everywhere else, I'm finding that large packets are for file transferring, and that gaming wants to use smaller packets.

More Smaller packets, vs Less Larger packets. and thats where I've been playing around.

I do note, that it seems to be less strain on the cpu with more smaller packets, than with packets larger than 512Bytes. but with the Larger packets, I see the server bandwidth stay closer to my 2Mbps upload. With the smaller packets, it seems to fluctuate alot more on the lesser than 2Mbps upload speed. Hard to judge, other than to keep testing different settings.

...Syn...

[GossamerSolid]

So then I should set my MaxMsgSend way lower.

Remember what I said, Bandwidth isn't the issue here. I'm trying to work around the hardware I'm provided with.

[VisceralSyn]

Do some research...

If you set your MaxMsgSend lower, you will limit the bandwidth the server uses. Find the smallest packet the server will send, then set the MaxMsgSend to use the maximum bandwidth you have.

Bandwidth shouldn't be an issue, because you have so much. But if you do not determine the optimal usage of said bandwidth, then its wasted.

With the guideline, that for TCP, 64byte packets are the smallest that it uses, before it determines its a fragmented packet. UDP *should* be able to go smaller, because it doesn't acknowledge.

I'm currently down to MaxMsgSend=4096; MaxSizeGuaranteed=64, and MaxSizeNonguaranteed=32; 2Mbps upload pipe to work with.

PacketsPerSecond Tool (http://www.ccievault.net/index.php/tools)
Try that to see what I'm talking about, PPS and MaxMsgSend, should be the same term.

...Syn...

[Wolffy.au]

Excellent thread guys! About time we got this stuff out in the open!

A fellow clan told me (and I confirmed) that JIP Lag and CPU stress are increased when you increase MinBandwidth/MaxBandwidth to those values.

Judging from watching stats while playing games, bandwidth per player ranges between 256kbps and 512kbps. Depending on your average missions, you want to use those numbers to work out the values. Currently I'm using the following:

language="English";
adapter=-1;
3D_Performance=1;
Resolution_Bpp=32;
Windowed=0;
MaxMsgSend=2048;
MaxSizeGuaranteed=1024;
MaxSizeNonguaranteed=64;
MinErrorToSend=0.0025;
MinBandwidth=4096000;
MaxBandwidth=16384000;

[GossamerSolid]

Excellent thread guys! About time we got this stuff out in the open!

A fellow clan told me (and I confirmed) that JIP Lag and CPU stress are increased when you increase MinBandwidth/MaxBandwidth to those values.

Judging from watching stats while playing games, bandwidth per player ranges between 256kbps and 512kbps. Depending on your average missions, you want to use those numbers to work out the values. Currently I'm using the following:

language="English";
adapter=-1;
3D_Performance=1;
Resolution_Bpp=32;
Windowed=0;
MaxMsgSend=2048;
MaxSizeGuaranteed=1024;
MaxSizeNonguaranteed=64;
MinErrorToSend=0.0025;
MinBandwidth=4096000;
MaxBandwidth=16384000;

I'll have to give these settings a try.

I don't really get much time to be fiddling with the bandwidth settings and testing so much.

[[KH]Jman]

Since you have posted the pipe is 100Mbps, why would you use 15Mbps as a minbandwidth? Even guessing for overhead should be between 88 and 92Mbps.
...Syn...

http://forums.bistudio.com/showpost.php?p=1673791&postcount=275

[VisceralSyn]

I'm not sure what your point is, Jman. If you do not define the bandwidth requirements, then the application does what it wants.

retarded?

Altering the MaxMsgSend, and the packet size, I managed to configure the server to use 100Mbps-LAN, and it sent me 12Mbps to just me alone.

Do what you will of others' information, or do your own research?

...Syn...

[[KH]Jman]

Chill out my friend, my point was to point the thread starter to the linked thread which contains much of the information required since this has been asked many, many times before.

[Defunkt]

Sorry but what you linked to was a single post that illustrates more of the typical guesswork on this subject. The type of experimentation and calculation that Syn is engaged in and which has the potential to yield maximal usage of available bandwidth (i.e. hopefully no more warping on a LAN) is long overdue.

Syn, I'd be very interested in seeing your cfg's in full (both for LAN and 2mbs Up).

[[KH]Jman]

Whatever and FYI it's not 'guess work'. Cross did some extensive testing on this subject, PM if you like. I'll leave you chaps to it.

[Wolffy.au]

The Kelly's Heroes configs are the benchmark for many server admins.

Re bandwidth, if your server cpu can take then load, then pump it up. Otherwise it will have the opposite effect.

[Defunkt]

Anyone can put up a webpage, it doesn't endow the information contained with any significance or accuracy and the fact that it's the only web page on the subject (and/or its contents frequently quoted or linked to) doesn't make it a benchmark, just the only one. Those numbers have never stacked up for me but I've never been willing to invest the time time required for a decent analysis and I'm very pleased somebody finally is. If Syn's config can require 12Mbps for a single LAN client then clearly saying "[...to utilise maxbandwidth] you'll need to have 120-130 people playing" betrays a significant gap in understanding, to infer the thread linked to already has all the answers is patently ludicrous.

[Taxman]

GET back on topic...

[VisceralSyn]

Smaller packet sizes, with the MinErrorToSend, has provided less stutter in MP. That last 100Mbps LAN test where I got 12Mbps was the last time I did a LAN test, all changes have been using the 2Mbps uppipe, and my brit gamin bud.

If I was to do a 100Mbps LAN config it would be:
MaxMsgSend=204800;
MaxSizeGuaranteed=64;
MaxSizeNonguaranteed=32;
MinBandwidth=92000000;
MaxBandwidth=100000000;
MinErrorToSend=0.001; // Any lower is just as bad as 0.1

Which should leave room for overhead. And, I might get to test that this weekend. For now, I'm trying to find where I read, setting packets as large as they can be is for file transferring, and Gaming should use much smaller packet sizes.

Then to determine, how much like TCP, UDP is.

Kelly's setting didn't make much sense to me either. And, no-one has gone thru and explained the config's tuning options. Which have the exact same information since Operation Flashpoint.

...Syn...

[Taxman]

This is great info guys, and im grateful (and im sure most admins are) but lets not Diss the one bunch of guys that up until now have been the only peeps to actually try and help us out (us= useless admins like myself) without them I would have still been on stock settings..

So thanks Kellys heroes..

going to try these new settings this evening.....
Thankyou very much for these...

[GossamerSolid]

Smaller packet sizes, with the MinErrorToSend, has provided less stutter in MP. That last 100Mbps LAN test where I got 12Mbps was the last time I did a LAN test, all changes have been using the 2Mbps uppipe, and my brit gamin bud.

If I was to do a 100Mbps LAN config it would be:
MaxMsgSend=204800;
MaxSizeGuaranteed=64;
MaxSizeNonguaranteed=32;
MinBandwidth=92000000;
MaxBandwidth=100000000;
MinErrorToSend=0.001; // Any lower is just as bad as 0.1

Which should leave room for overhead. And, I might get to test that this weekend. For now, I'm trying to find where I read, setting packets as large as they can be is for file transferring, and Gaming should use much smaller packet sizes.

Then to determine, how much like TCP, UDP is.

Kelly's setting didn't make much sense to me either. And, no-one has gone thru and explained the config's tuning options. Which have the exact same information since Operation Flashpoint.

...Syn...


So with these settings it would technically use all of 100mbps?

[VisceralSyn]

Yes, 204,800 64byte packets equals 100Mbps. Do note, thats not counting for overhead, 100Mbps is impossible to reach on a 100Mbps LAN connection.

After some more scouring the internet for more information on UDP, I think i can nail down the stuttery AI. UDP, while connectionless, also does not guarantee delivery or order. So if a client receives packets out of order, it drops it. Since unit location updates fall under MaxSizeNonguaranteed, I have commented the line out. In case its buffering to fill the packet, instead of just sending the information. Currently testing my server with:

MaxMsgSend=8192;
MaxSizeGuaranteed=32;
// MaxSizeNonguaranteed=;
MinBandwidth=2000000;
MaxBandwidth=2200000;
MinErrorToSend=0.001;

I got to fire it up this morning, seems to be less CPU utilization on Warfare BE 2.066. Will monitor CPU usage during the day, when I get off, will try to get some clients in it. To double check AI stutter, which I noted, was amazingly less when I started the mission.

...Syn...

[Taxman]

could you tell me.... If im not using ANY AI would these settings improve our server ?

[VisceralSyn]

You would need to set your server, in accordance with the bandwidth availible to your server.

If there is little to no AI, then scripts would be the next resource to bog down the processor. If you have the stutter of other units, and you have the MinErrorToSend=0.001; then the next parameter to start working with is the packetsize regarding unit updates. MaxSizeNonguaranteed. AI units are not the only cause of processor usage, from what I'm finding the stutter everyone hates is a cause of UDP, dropping packets that are out of order. But AI units aren't the only thing that needs updates.

Test of commenting out the MaxSizeNonguaranteed, so far CPU usage is down, just over 3 hours later.

...Syn...

[Taxman]

I have run the Benchmark mission (from the other thread) on both of your latest cfg's and I am getting lower server fps than my original KH setup....

will give it a go this evening on Pvp Missions to see how it plays out buddy... will update tomorrow.

(regards to cpu, no noticeble diff with our server)

[Taxman]

MaxMsgSend=204800;
MaxSizeGuaranteed=64;
MaxSizeNonguaranteed=32;
MinBandwidth=92000000;
MaxBandwidth=100000000;
MinErrorToSend=0.001

Tried those settings.... PvP with only about 7 peeps on...Low cpu yes............ Desync on server was through the roof.. everyone had around 10000.... unplayable matey... had to revert to original

---------- Post added at 23:11 ---------- Previous post was at 23:10 ----------

I guess, your settings maybe ok for lan but sadly not dedicated server
least not the ones I tested

[Crimson_Raptor]

What if it didn't work because your host doesn't "really" have a 100mbit connection? Unless you host your own, I've found that many server hosts out there usually have two or 3 different games on one PC, but they're all sharing not only system resources, but the 100mbps connection as well. Some are very shifty...

[VisceralSyn]

I guess, your settings maybe ok for lan but sadly not dedicated server
least not the ones I tested


I've got no idea what causes De-Sync, other than a mission initializing. during a session it should not de-sync. I suggest a lower MaxMsgSend, and higher MaxSizeGuaranteed.

MaxMsgSend=102400;
MaxSizeGuaranteed=128;

I do my testing with both client PC, and Dedicated server pc on a 1Gbps LAN. With MaxMsgSend setting, and MaxBandwidth I set my server for the internet connection, which is 2.2Mbps ( I only set for 2Mbps ). Again, you must configure your server for its bandwidth or like the above statement, makes little sense.

...Syn...

---------- Post added at 06:53 PM ---------- Previous post was at 06:50 PM ----------


What if it didn't work because your host doesn't "really" have a 100mbit connection? Unless you host your own, I've found that many server hosts out there usually have two or 3 different games on one PC, but they're all sharing not only system resources, but the 100mbps connection as well. Some are very shifty...


Go for a more conservative Max/MinBandwidth setting, like 20 or 25Mbps. Either of which are still significantly more than my piddly 2.2Mbps upload speed.

...Syn...

[SWAT_BigBear]

What if it didn't work because your host doesn't "really" have a 100mbit connection? Unless you host your own, I've found that many server hosts out there usually have two or 3 different games on one PC, but they're all sharing not only system resources, but the 100mbps connection as well. Some are very shifty...
ServerBeach (my last dedi box) is a Peer1 dual OC192 fatpipe.
I had options of 10mbit port or the 100mbit port. Those are just the network to the fatpipe.
I actually had 28MB up, 6MB down with the 100mbit. That comes with the basic packages. Now, of course you can pay $1500+ extra per month and get the solid up/dn speeds of the fatpipe. But who would pay that for gaming?

[Taxman]

What if it didn't work because your host doesn't "really" have a 100mbit connection? Unless you host your own, I've found that many server hosts out there usually have two or 3 different games on one PC, but they're all sharing not only system resources, but the 100mbps connection as well. Some are very shifty...


Buddy... honestly what makes you think my host doesnt give me what I pay for.... jeez.. IT didnt work IT caused desync and this has nothing to do with my 100mbit connection which btw is on the killercreations network,,, pretty good network.

I have a Dedicated server, The whole system belongs to us....(rented :) ) but I have FULL access to desktop, and ALL four cores,


Gossamer, have you tried these settings buddy ?

Now I will try some of the other suggestions.... not sure what the causes of the desync are myself buddy BUT it isnt mission specific on this occasion, same mission other stock KH settings no desync.. ..
I will give your other figures a shot and let you know... hopefully we can get to the root of the problems, wont know til prob sunday though...

THank you for your help bud.

---------- Post added at 08:42 ---------- Previous post was at 08:38 ----------


ServerBeach (my last dedi box) is a Peer1 dual OC92 fatpipe.
I had options of 10mbit port or the 100mbit port. Those are just the network to the fatpipe.
I actually had 28MB up, 6MB down with the 100mbit. That comes with the basic packages. Now, of course you can pay $1500+ extra per month and get the solid up/dn speeds of the fatpipe. But who would pay that for gaming?

Im sure mine is around 50 up and 50 down......

[VisceralSyn]

Really hard to say, if commenting out the MaxSizeNonguaranteed line is a benefit. Still see the AI stutter a bit around the 1km mark. But its not consistent.

I got a second local client, and my client lined up for weekend testing. Guess we'll see how it goes saturday.

...Syn...

[GossamerSolid]

I haven't gotten to test any of these very well as I've only had like 4 people on to test with.

I know my connection is capable of handling over 100mbps. They just did upgrades a while back:

http://www.speedtest.net/result/1121376835.png (http://www.speedtest.net)

[Taxman]

Jeeeeeeeeeez goss thats one hell of a connection lol

[cri74]

*following this thread - allso trying to tweak the settings :)

Allso a thread of intrest on this subject (http://forums.bistudio.com/showthread.php?t=111902)

[GossamerSolid]

Jeeeeeeeeeez goss thats one hell of a connection lol

Yes...

Yes it is :)

Good Ole' InterNAP and their strong networks.

[Taxman]

http://www.speedtest.net/result/1121731689.png (http://www.speedtest.net)

And I was happy with this lol....

You guys accross the water have some serious connections... very jealous bud..

[GossamerSolid]

Hey Syn, I tried these settings:

MaxMsgSend=8192;
MaxSizeGuaranteed=32;
MinBandwidth=2000000;
MaxBandwidth=2200000;
MinErrorToSend=0.001;

And server performance was the same, but Client-side performance was way better while on the server with more units on the map. Could Server bandwidth settings affect Client FPS?

EDIT: Just realized the bandwidth was set for 2.0 - 2.2 mbps. I've set it to 20 - 22mbps for a future test.


Here's our findings from testing with AI on servers if anybody is interested: http://forums.bistudio.com/showthread.php?p=1839552#post1839552

[VisceralSyn]

I'm getting something similiar to that on 100Mbps LAN. The server FPS is very low, but still in double digits. Its ok for the the other LAN client, but mine is stuttery in and near towns.

I'm going to leave the MaxSizeNonguarantee'd commented out, and raise MaxSizeGuaranteed and lower MaxMsgSend, until it stops, or I see no change.

since i'm currently doing:
MaxMsgSend=102400;
MaxSizeGuaranteed=128;

going to:
MaxMsgSend=51200;
MaxSizeGuaranteed=256;

Again, for 100Mbps LAN. Weirdly enuff the server monitor has yet to register over 256Kbps out for both clients.

I'd also be very interested in that stripped down, possible benchmark-mission.

...Syn...

[VisceralSyn]

HUGE...

Do not put the dedicated server bandwidth tuning lines in a client's *.cfg. It does cause a ton of additional stutter, and likely the cause of all the stutter I got near towns.

2 hours in Benny 2.066 and fps at 25, removed a few ai teams, 12v12 insteada 16v16.

Not doing the 100Mbps LAN config, back to 2Mbps config, I have 2 LAN, and 1 Internet clients. Each seem to be getting about 256Kbps each, server bandwidth is staying right around 768Kbps.

...Syn...

[Taxman]

what was the server fps?

Btw Not really understanding the post regarding the clients *.cfg ? could you clarify please..

[brit~XR]

I'm not sure what your point is, Jman. If you do not define the bandwidth requirements, then the application does what it wants.

retarded?

Altering the MaxMsgSend, and the packet size, I managed to configure the server to use 100Mbps-LAN, and it sent me 12Mbps to just me alone.

Do what you will of others' information, or do your own research?

...Syn...


I saw your settings and to be honest there just awful. Anyone who can recommend using MaxMsgSend=204800 has no idear. Thats just insane. Even 2048 would cause huge desync problems.

1024 would be my max limit if its for 64+ players pvp

You think sending 12Mbps to you alone is good? Why try give a server more bandwidth then it needs. Just causes problems. My ofp res server and arma server use around 4.5mbit for around 32 players. If i was to set a 32 player cti server to use 20mbit like some pvp servers use for 100 players then id get huge desync problems. Have you actually tested your settings with atleast 32 players? Any settings can appear to run fine with few players.

I just find it hard to beleave them settings would work on any populated server without desync issues

[cri74]

Have you actually tested your settings with atleast 32 players? Any settings can appear to run fine with few players.
-This is true

We are trying some of the settings back an forth now. The funny thing is the setting thats we are trying now. Server FPS at around 10-15 from the start with around 20 ppl online and it seems to hold 10 FPS untill the game finish (2-3hour sessions with more or less players thrughout the session), but studdereing and rubberbanding is much less... allso desync is much less. Takistan seems to have more problems than Chernarus FPS whize wich is really strange to me, since i think of Takistan to be less demanding, but this might allso be scripting or mission build. Or it might be that the new settings is actually holding back information ? (kind of like a racing motor in a Lada :p) I guess (for us maybe) the FPS counter might not be telling the whole truth and cant be the only parameter to look at when tuning the server. Is there a good way to use the monitor command and if so would anyone care to give me some more insight pls?

MaxMsgSend=51200;
MaxSizeGuaranteed=256;
MaxSizeNonguaranteed=256;
MinErrorToSend=0.001;
MaxCustomFileSize=0;
Windowed=0;

The MaxMsgSend is really high now as you see and some other settings might not be good, but we are experimenting and trying to learn how it effects our server since it seems everyone has their own way of doing it :)

This and the other threads linked is really helpfull so thanks for creating and posting, i'm learning :)


Wish we could get a BIS developer here to explain more

[VisceralSyn]

I saw your settings and to be honest there just awful. Anyone who can recommend using MaxMsgSend=204800 has no idear.
Thats just insane. Even 2048 would cause huge desync problems.


Whats an 'idear' ?

Since math is elusive, let me put something down on:
204800 * 64Byte (MaxSizeGuaranteed) = 100Mbps

When the packet's size is adjustable, you have complete control of how much bandwidth you can use. Sounds like a great idea to me! If some people are smart enough to use math.



1024 would be my max limit if its for 64+ players pvp


Thats nice, thats your choice




You think sending 12Mbps to you alone is good? Why try give a server more bandwidth then it needs. Just causes problems. My ofp res server and arma server use around 4.5mbit for around 32 players. If i was to set a 32 player cti server to use 20mbit like some pvp servers use for 100 players then id get huge desync problems. Have you actually tested your settings with atleast 32 players? Any settings can appear to run fine with few players.

I just find it hard to beleave them settings would work on any populated server without desync issues


If during research you discover that the server can and will use more bandwidth during, if configured to do so. Obviously, you did no such research.

I'm at least trying other things, and not guessing what it should do. If I knew 32 people who has this game, I'm pretty sure we would put money together to rent a dedicated server on a 100Mbps internet connection, and not my piddly 2.2Mbps-UP Cable connection.

I will add, the testing done yesterday with the MaxSizeGuaranteed set to 32, I ended up restarting, at 64, and MaxMsgSend to 104200, which again, equals 100Mbps, for the math-challenged peoples. Removed bandwidth in general as part of the serving problem. 2 Lan clients recieved a steady 256Kbps. Later switched to Internet, 2Mbps up. Which also means mathematically less MaxMsgSend.... and got similiar bandwidth usage, approx. 256Kbps per client, oddly enough on internet the bandwidth fluctuated more than on LAN.

...Syn...

[GossamerSolid]

I guess (for us maybe) the FPS counter might not be telling the whole truth and cant be the only parameter to look at when tuning the server.

If you are using warfare, it's as accurate as can be. It's made with this command: http://community.bistudio.com/wiki/diag_fps.

If you type #monitor you will see the same results

[VisceralSyn]

Commenting out MaxSizeNonguaranteed provides the best solution to AI Stutter. It is positively less.

...Syn...

[cri74]

Yeah - but fiddeling around with the settings seems not only FPS is a value that describes how players experience the game oddly enugh.


AI studdering


From
MaxMsgSend=1024;
MaxSizeGuaranteed=1024;
MaxSizeNonguaranteed=384;
MinErrorToSend=0.0049999999

To
MaxMsgSend=1024;
MaxSizeGuaranteed=256;
MaxSizeNonGuaranteed=128;
MinErrorToSend=0.001;

Allso gave a lot less AI studdering, but created desync

[Taxman]

May I just add that we are all trying to obtain the same thing,,,, a SATISFACTORY solution to our Server configs.....

Please, can you all refrain from these sarcastic comments.. it is not doing this thread any good... Syn, I appreciate what you are trying to do BUT you really need some people skills buddy.... Brit XR also has some valid points (maybe he should have worded it better and not pissed you off) and im sure if we ALL work together we can come up with some good configs that suit ALL needs....

If you need some info about running a 32 player server then all you need to do is ask,, you dont need 32 more friends, we can help.

One thing about this thread proves, is that we need each other ffs.. so lets just lighten up ...

[cri74]

I will be the first to admit....i threw away my math books a long time ago and started playing computer games instead - so all this might be over my head :p If its all down to math i am happy we have a math wiz here explaining things to me :)

[GossamerSolid]

Expanding on what Taxman said,

It's also hard to determine what the "Best" config is as we all have different connections and different hardware/software running on our boxes.

Plus we are running different missions in some cases as well.

[djczing]

I might suggest that there is another layer that is important to these settings which hasnt been disucussed -- thats the MTU (Maximum Transfer Unit) of your LAN and/or internet connection.

MTU 1500 is the norm, and apparently optimal when running over a DSL or cable modem (3-6Mbps down).

One of the factors which should be considered is how well the data/msg sizes fit in a single eithernet frame at 1500 MTU (or a higher MTU if you decide to change that).

Msgs sent which are not a multiple of MTU will necessarily be fragmented, and could take 2 ethernet frames to complete, rather than just 1.

The MTU of the interface does include header data, so adjusting your msg sizes to the actual data /payload/ size of your ethernet frame should be the most efficient way of using the underlying network. Any multiple of your (MTU size - header data = payload size) should be the most efficient use of your underlying network.

I run a small server on a 1Gbps LAN, and have been playing with the settings a bit. With one of 2 users the #monitor data indicates very low network useage, and thus it wouldnt seem that with a small number of users that network bandwidth/useage is a factor.

The minerrortosend fixed my AI rubber-banding issues for the most part, as I have it set to .005.

I think the rest of the issue is simply CPU load, and how long the CPU takes to make a complete game update cycle. The more AI and objects it has to deal with, the harder it is to push through updates on them all.

Having the network settings set aggressively (like with a high maxmsgsend) might look good at first, but it also might take CPU cycles away from game processing.

Tuning is never straight-forward. Its always a balance.

Enjoying the thread. Thanks all for your work.

[cri74]

After testing multiple settings this weekend with full servers (mission Gossamers warfare) we have found what we consider to be close to optimal settings for Takistan. Snipers and regular classes report a much more fluid and better experience. The same settings in Chernarus however will not produce the same result and we will try tweaking to find the same optimal experience for that island and put this on a separate server. Server 1 and server 2 / Takisatan and Chernarus, compared to before when we had a negible settings that could run on both islands with what seems to be a inferior experience compared to splitting the servers by islands.

[Birdseye]

MTU is interesting, besides that, maybe the netsh commands have some influence?



(With certain hardware, these features completely mess up network traffic :P):
(Run from Elevated Command prompt)

netsh int tcp set global rss=disabled
netsh int tcp set global autotuninglevel=disabled
netsh int ipv4 set global taskoffload=disabled

optionally you could try this one too:

netsh int tcp set global chimney=disabled

You can review the default settings if you like with these commands:

netsh int tcp show global
netsh int ipv4 show global

[=WFL= Sgt Bilko]

Commenting out MaxSizeNonguaranteed provides the best solution to AI Stutter. It is positively less.

...Syn...Wouldn't remarking MaxSizeNonguaranteed be the same as setting it to 256 (default according to wiki)?

[VisceralSyn]

Execellent question, I wish I had the answer for. Commenting it out, produced the least amount of AI running stutter, 2nd best was 24~32.

@<hidden>, MTU is ISP limited, ethernet is 1500. I tested mine with a windows machine:


ping xxx.xxx.xxx.xxx -f -l 1472


Produced 0 fragmented packets, with a ping to my ISP's homepage, Comcast.com. Sadly, Gigabit-Ethernet's jumbo-frames, cannot be used over the internet.

Still not sure of, on the whole smaller vs larger packets deal, if ethernet's fragmenting packets larger than the MTU, or padding packets smaller than the MTU is actually happening, with UDP.

...Syn...


...Syn...

[SWAT_BigBear]

In the OP, he stated:


Here's my current server config:
MinBandwidth=15000000;
MaxBandwidth=100000000;
MaxMsgSend=1024;
MaxSizeGuaranteed=1024;
MaxSizeNonguaranteed=64;
MinErrorToSend=0.0024999999;
MaxCustomFileSize=0;

From day one of this thread, I'm wondering if he meant in his basic.cfg or arma2.cfg?
Or, do I misunderstand the WIKI (http://community.bistudio.com/wiki/basic.cfg) ?
Are we able to put them in the server.cfg?

[GossamerSolid]

I meant the arma2.cfg

You can't put that stuff in the server.cfg

[stk2008]

This is all over my head.

I am trying to tweak me server config but I have no idea what to set it all to :(

Here is my broadband performance results

http://www.speedtest.net/result/1169590717.png (http://www.speedtest.net)

The max amount of players that will connect I cant see going over 5.

I am using real crappy hardware at the moment for me dedicated server but I hope to change all that soon.

Oh and I of course connect to my dedicated server VIA LAN and my mates over the internet.

What would be a good starting point for my config file?.

MinBandwidth=
MaxBandwidth=
MaxMsgSend=
MaxSizeGuaranteed=
MaxSizeNonguaranteed=
MinErrorToSend=
MaxCustomFileSize=

Thanks very much in advance

[SWAT_BigBear]

stk2008, you should 1st try not using any settings. If only 5 will be playing, I doubt a basic.cfg will be needed. Your internet speed should be ok for 5.

[Brute]

AI stuttering actually seems to become more apparent when changing MinErrorToSend=0.005 to 0.001?

It would be quite nice to have BI developer actually join in the discussion and help.

[Brute]

I have a server with the following specifications:

Intel Quad Core Xeon 2.66GHz (x2)
8GB RAM
100mbps line

While running a Warfare BE map I can get quite a considerable amount of warping. I have been adjusting the values in my basic.cfg however it seems that changing basically any of the values seems to cause even more warping. I need some help with exactly what to set. I'm aiming for between 48 and 64 players.

[GossamerSolid]

I have a server with the following specifications:

Intel Quad Core Xeon 2.66GHz (x2)
8GB RAM
100mbps line

While running a Warfare BE map I can get quite a considerable amount of warping. I have been adjusting the values in my basic.cfg however it seems that changing basically any of the values seems to cause even more warping. I need some help with exactly what to set. I'm aiming for between 48 and 64 players.

You won't get 48 to 64 players with that server. A major issue with ArmA 2 dedicated servers is that it doesn't matter how many cores you have (above 2), it relies on the core clock.

I don't know if I'd even put 32 players on that server.

(By player count, I'm talking warfare-based).

[Brute]

It seems to be running fine at 64 so I'm not so sure your advice is proving true, thank you though.

[VisceralSyn]

If the server FPS is 15 or less, according to the readme for the dedicated server, it would not be running fine.

...Syn...

[.kju [PvPscene]]

above 10 FPS is completely fine
lower than 5 gets problematic
less than 3 causes serious issues

[NoBrainer]

Yeah - but fiddeling around with the settings seems not only FPS is a value that describes how players experience the game oddly enugh.


AI studdering


From
MaxMsgSend=1024;
MaxSizeGuaranteed=1024;
MaxSizeNonguaranteed=384;
MinErrorToSend=0.0049999999

To
MaxMsgSend=1024;
MaxSizeGuaranteed=256;
MaxSizeNonGuaranteed=128;
MinErrorToSend=0.001;

Allso gave a lot less AI studdering, but created desync

What are you using now?
Have you found a good solution?

[GossamerSolid]

above 10 FPS is completely fine
lower than 5 gets problematic
less than 3 causes serious issues

I'd say lower than 10 gets problematic and lower than 5 causes serious issues. Under 20 and you start to notice more AI warping.

[Brute]

It seems to remain above 30 fps, I have had eight hour games with 30+ players and the server has never gone below 30 fps so I guess the server hardware is doing just fine :). I have been considering getting a couple of X5690s though so I'm sure that will improve things and hopefully allow me to run 64+.

[dorkirt]

Please, aynone know a good config for a

Quad Core 3 Ghz, 4GB RAM and
http://www.speedtest.net/result/1284333938.png

Max players 16 ?

Is this OK ?

MinBandwidth=2000000;
MaxBandwidth=5000000;
MaxMsgSend=512;
MaxSizeGuaranteed=1024;
MaxSizeNonguaranteed=64;

[Brute]

GossamerSolid, how many players do you think the following server could handle?

Intel Xeon X5690 (3.46GHz (3.73GHz Max Turbo Frequency) 6C/12T) (x2)
24GB RAM
1 GBps

[GossamerSolid]

GossamerSolid, how many players do you think the following server could handle?

Intel Xeon X5690 (3.46GHz (3.73GHz Max Turbo Frequency) 6C/12T) (x2)
24GB RAM
1 GBps

depends on what mission you intend to play.

[Brute]

Warfare BE

[larsiano]

@<hidden>: Its like asking how fast the starship Enterprice goos to a NAVY captain, and the only answer is "unknown". This is as fast as Intel will go without paying more then 100K for a machine so its the fastest of the moment not allot of server admin have experienced this kinda speed.
So in the end the server speed wont be the issue but maybe the network speed will crumble sooner ...but who knows? Why dont you test it yourself with a 100 player mission? I'dd love to know the answer...

[Brute]

Iarsiano, I realise that not many, if any, server administrators have had access to such a powerful machine to run Warfare BE on. I was simply asking for his opinion, I thought he may have used a server which had a similar clock speed, since apparently the server client is restricted to only using two cores, and could therefore advise me on an approximate player count to begin experimenting at.

The server won't be installed till next week however, so I will have to wait to see what it can handle. I doubt the bandwidth will prove to be the bottleneck though. I would very much like to find a way to force the server client to distribute the AI operations over all available cores since then I imagine you could quite easily have 128+ players.

I think I will begin testing at 64 players and see how things go. Keep an eye out for the server in the next week or two :).

[Brute]

Setting the min bandwidth to 75% of a 100mbps line actually decreases performance, it's ridiculous.

[Overlord]

Interesting reading indeed. Many of you runs tests on highspeed LAN's, and using those tests to configure online servers.
I guess there are some issues who have to be taken in consideration;

LAN games normally don't use HOPs for clients, how will an online server be affected by clients going over e.g. 25-30?
How will client performance affect server? If a client have "bad" internet connection and/or low-end system? I remember this issue from good ol' OFP, if certain players connected, some missions got un-playable (no addon related).

I remember reading something about if you cut out server settings (BW settings), the server would calculate the best settings. This will of course create lot of useless CPU usage, but my question is (if true): does the server save these setting "on the fly"? And use those settings as later on, so in time you would get a fully configurated server?

[Rexxenexx]

It would be nice to have a program (or website) that would test bandwidth (like to one of those speedtest sites) and then write a config file. I'm sure someone can think of a close enough algorithm to at least get an idea of what the settings should be. That way we don't have to keep guessing.

EDIT: Maybe even just a site that has text entry for speedtest download and upload bandwidth, and also a dropdown box for bandwidth % you want to use. [100%|90%|75%|50%|25%] Then when you press Submit it returns the config file.

I'll write it in PHP if you guys give me an algorithm. x = download speed, y = upload speed, % = users selected bandwith percent.
The return would be:


MinBandwidth=____________;
MaxBandwidth=____________;
MaxMsgSend=____________;
MaxSizeGuaranteed=____________;
MaxSizeNonguaranteed=____________;
MinErrorToSend=____________;

[Overlord]

OK, try with 50/50 Mbps, dunno about limit clients, ain't it good if connected clients share available bandwidth?