DCS World and NetCode

[Rikus]

Now that with DCS World, there will be more options to choose what you want to fly, we will see missions with more players.

 

Will the netcode improve at level of BMS or IL2?

 

Greetings

[TAGERT]

That is a good question!

 

And this problem has been around since LockOn.. So I fear they are not going to fix it any time soon

 

Nothing pops your immersion bubble quicker than seeing a plane sitting on the runway one second.. warping to 10,000ft the next second.. than back onto the runway the next second

 

As for jet missle kills, ie BVR kill it is not a real big issue.. Only because you can not 'see' a plane warping at BVR!

 

But now with up close and personal gun kills type of fighting the P51 brings to the table

 

The net code is going to have to be improved!

 

You just can not have planes warping around as your lining up your gun sight

 

If they dont fix it WWII gun kill stuff will never get off the ground (pun intended)

 

Except for when it warps off the ground!

Edited June 3, 2012 by TAGERT

[Essah]

Now that with DCS World, there will be more options to choose what you want to fly, we will see missions with more players.

 

Will the netcode improve at level of BMS or IL2?

 

Greetings

 

you must be kidding me. BMS Netcode is terrible!

however I agree that a netcode improvement is needed. I would rather see one aircraft less and have the netcode improved, than the alternative.

[Blaze]

I'm sorry but after playing the Lock-On and DCS series since it's creation in 2004 I think it still has one of the better netcodes (if not the best) in the sim community, even to this day. :noexpression:

Edited June 3, 2012 by Blaze

[Dimebag]

I'm sorry but after playing the Lock-On and DCS series since it's creation in 2004 I think it still has one of the better netcodes in the sim community, even to this day. :noexpression:

 

Would have to agree there, the formation flying community is a testament to that.

[Kenan]

 

Will the netcode improve at level of BMS or IL2?

 

Greetings

 

BMS has a worst netcode ever..I can host only me and my buddie in BMS, yet, in FC2 I can host up to 6 players easy (And I have a crappy upload).

[Leto]

I don't know what this thread is about. I flew IL2 from first beta till last year and state: netcode is much better in DCS. Formation flying is a blast in DCS.

[Panzertard]

What is "better netcode"?

In reality you need to specify what is broken - why it is broken and how it can be improved.

-----

Data quality & world precision

Inability to host for more planes can mean it's just more complex - more data to synchronize - not that it's crappy.

Of course, if you bring more airplanes into a server - and retain the same bandwidth, then you may expect worse precision (data quality) for all the elements in the world. Cost of data-quality vs data-quantity.

 

So what is 'better netcode'? Less data & worse precision? Or better precision and less clients & objects?

Or perhaps 'better netcode' really means; Get better lines if you expect more planes and the same precision for the environment?

 

Network reliability

So what to improve? UDP reliability?

This is one of the major factors which is what causes 'spikes' and lagging. Loss of UDP packages reaching your PC.

And when they finally do arrive - the enemy is in another place.

 

Now you could start changing UDP packages with TCP. But then you might realize that the latency in a dogfight would be intolerable due to the increased headers and extra checks with this kind of change. The bandit would be in another place by the time you press the trigger.

No, this may indeed not be a good change.

 

Now instead you could ask your opponent to get a better computer & router or change his ISP, so you have a better chance of killing him.

Nah, j/k - but you get my point. ;)

 

2mbps up/downlink. FC / DCS. 14-16 players in a rich environment, 500-1000 units. Good precision, very little lag / loss of UDP.

I can't see how this is bad.

 

I've flown dogfights with pilots from the other side of the world having 400-600 ms ping. Still great precision.

I can't see how to improve on this.

 

UDP packages will get lost. You can't change this fact unfortunately.

 

PS, I've hosted missions with 10-12 players, with 2000+ units.

Unfortunately the pilots instead complained about how their computers started to melt.

Not that the net was unreliable.

Edited June 3, 2012 by Panzertard

[Bushmanni]

Yes, DCS netcode is the best I have seen in sims. The only improvement could be better interpolation when the packets get lost so as to prevent physically impossible warping like aircraft momentarily rising 10m in the air while taxiing.

[sorcer3r]

I guess netcode isnt a problem (but we will see how it runs with more clients when FC3 is released). More important is the server/client stability.

 

Only unpleasant issue I had (1.1.1.0/1) was about the incorrect flightpath of some missiles (against client Ka-50.) But maybe solved since 1.1.2.0.

[Essah]

I realise how frustrating it must be to listen to people like us that have no idea what we're asking for. So thanks for that clarification Panzertard

 

I actually don't remember seeing the massive warping in DCS that I remember from FC1 and 2. So either i've been playing with people with better connections or something has been improved.

Like other people said. We probably shouldn't really complain since we got some of the best there is.

 

what I personally meant with "better netcode" is a general overall improvement, not at the cost of anything. Of course this is probably a pipe dream, but anything else would not be worth striving for, and Like I said. I actually haven't seen those big lag spikes lately so I'm not gonna complain.

[c0ff]

Yes, DCS netcode is the best I have seen in sims. The only improvement could be better interpolation when the packets get lost so as to prevent physically impossible warping like aircraft momentarily rising 10m in the air while taxiing.

 

In DCSW I tweaked the code to be less sensitive to spikes in ping. It should noticably reduce warping.

[c0ff]

what I personally meant with "better netcode" is a general overall improvement, not at the cost of anything. Of course this is probably a pipe dream, but anything else would not be worth striving for, and Like I said.

DCS (and before it, LOMAC and FC) are unique products. There're no recipes how to do networking in a similar sim/game environment.

The code for interpolating of aircrafts's position was developed by me and Yo-Yo specially for LOMAC. Now it's used for vehicles as well. As with all engineering solutions it's not perfect, but we tweak it as we learn, and feel that the benefits of our approach far outweight the drawbacks.

[Panzertard]

I realise how frustrating it must be to listen to people like us that have no idea what we're asking for. So thanks for that clarification Panzertard

 

...

 

what I personally meant with "better netcode" is a general overall improvement, not at the cost of anything. Of course this is probably a pipe dream, but anything else would not be worth striving for, and Like I said. I actually haven't seen those big lag spikes lately so I'm not gonna complain.

 

not at all, sometimes we just try to put it in perspective and feed you with a bit of info for further discussion.

 

Besides, see what you managed to do, rip c0ff out of his zombie state of programming, allowing him some fresh air on the forums. :D

Thanks for the update c0ff. :)

[Bushmanni]

In DCSW I tweaked the code to be less sensitive to spikes in ping. It should noticably reduce warping.

 

Cool. :thumbup:

[Kuky]

In DCSW I tweaked the code to be less sensitive to spikes in ping. It should noticably reduce warping.

 

Thanks c0ff, fingers crossed here as warping has really pi**ed me off when flying MP

[winchesterdelta1]

Maybe improvement in interpolation like Arma 2 did??? But i dont know how netcode works so disregard if this is not possible for DCS.

 

(sorry.. ninja'd by a dev)

Edited June 6, 2012 by winchesterdelta1

[Kuky]

ArmA/ArmA2 is nothing special in net coding as there is plenty of pretty much constant warping there also, even more so than LO/FCx/DCS

[winchesterdelta1]

ArmA/ArmA2 is nothing special in net coding as there is plenty of pretty much constant warping there also, even more so than LO/FCx/DCS

 

Sorry than you never installed the 1.60 patch. SO not constan warping anymore. There is barely any warping now. Unless you code your mission wrong or spawn 1000+ units. So yes there is something special with that netcode.

[nemises]

warping is a function of latency and interpolation...

 

if you are in Australia , with a round trip delay of 300 msec to a US server, but the server tick update time is 100 msec (guesswork numbers by the way) , then the server will have to update your client position using interpolation from updates 300 msec in the past for 2 server ticks ahead.

 

If you are flying straight and at a constant speed, then the interpolation will be pretty accurate (the server will say...300Msec ago you were at X location heading in Y direction at Z speed, therefore your most likely position at this time is X+(y*Z) (obviously grossly simplified!)

 

If your position and heading is changing very rapidly though, then the interpolated data will be some distance from the true data.

 

to overcome this, the interpolation will probably use some sort of curve to differentiate your true position from your interpolated position...this makes it look a bit nicer for the other players on the server for eg...but the by-product of using a calculated curve for position interpolation is some times the curve could reach mathematical boundaries , and cause the interpolated positions (the positions that are used between true position updates) to be wildly off the mark, such as when you see a client jump from Taxiing, to 10000ft and then taxiing again...

the alternative though, is to just use straight line interpolation, which is literally connect the dot's..but this would look very VERY ugly to other clients...all clients would be rapidly changing positions at the rate of server ticks + the latency with no smooth transition in between.

 

..sorry, long "answer" .. but all said and done, the higher your latency to the server, the more jerkiness other clients will see from your client ...

 

Other games can get around this by either having a very narrow server tick (however this is at the expense of processing availability, and requires clients to have a very low latency)...or by having a very wide update window (which would only really work in a sailing simulator where client changes are not time sensitive to the 1/100 msec)

 

all imho of course!

[Fish]

warping is a function of latency and interpolation...

 

all imho of course!

 

Great post. cheers

[TAGERT]

Data quality & world precision

Inability to host for more planes can mean it's just more complex - more data to synchronize - not that it's crappy.

 

Of course, if you bring more airplanes into a server - and retain the same bandwidth, then you may expect worse precision (data quality) for all the elements in the world. Cost of data-quality vs data-quantity.

I have seen this what i described on all different sorts of servers, dedicated with many players and single PC hosted with just a few players, each with solid pings.

 

And before anyone says the common denominator is my PC and connection, allow me to point out that I am not the only one witnessing these events. Other in game have report the same anomalies.

 

And before anyone says the common denominator is the HOST, note above where I stated I have seen this what i described on all different sorts of servers, dedicated with many players and single PC hosted with just a few players, each with solid pings.

 

An added note, I have been flying online for 20+ years.. Started back in the day with Air Warrior, before the Internet, when the Genie network supplied the connections (modem) and it only cost $12/hr to play!

 

So, I have seen my fare share of warping senarios and can typically spot the warping due to the one guy with the bad connection vs. something that is common to all.

 

So what is 'better netcode'? Less data & worse precision? Or better precision and less clients & objects?

In either case, there will be times that the PC program has to estimate (predict aka interpolate) the location of the object in the 3rd based on the previous locations. Which is something the PC program has to do in between data point. This is where I suspect the problem is.. Or should I say 'specify what is broken'. What leads me to belive this is the way the program (DCS) recovers from a 'bad' prediction.

 

For example, with regards to the example I provided, where a plan is sitting on the run way or in the process of taxi on the run way.

 

1) Data packet n: it is where I can see it, on my RHS about 20 ft away

2) Data packet between n and n+1: it is where I can see it, on my RHS about 20.5 ft away

3) Data packet n+1: it is where I can see it, on my RHS about 21 ft away

4) Data packet between n+1 and n+1: it 5k ft above me

5) Data packet n+2: it is where I can see it, on my RHS about 22 ft away

 

The fact that the object location ultimtally recovers from the warp implies that once it receives 'real' object locations it knews where to render the object in the 3D world..

 

Where as the estimated (predicted aka interpolated) location code failed to predict a valid object location.. Which resulted in the object location being 5kft above me. Note my description above was only n to n+2 where as in reality there are many more data packets being collected in a second.. Which is why you can see the plane 'move' (being rendered) between ground zero and several points along the way up to 5kft. What this implies is the FIRST bad estimated (predicted aka interpolated) location did not recover immediately upon receiving a valid packet because that value is being filtered and thus takes it awhile to move through the filter before it fully recovers. At least that is how it acts, it could be caused by something entirely different. Which I suspect is the case, because it is pretty simple to add code to detect bad values like that and keep them from being used and thus never upsetting the filter. I just give that example for those that have worked with code like that to give them an idea of what it looks like is going on.

 

Or perhaps 'better netcode' really means; Get better lines if you expect more planes and the same precision for the environment?

As noted above I have seen this what i described on all different sorts of servers, dedicated with many players and single PC hosted with just a few players, each with solid pings.

 

Network reliability

So what to improve? UDP reliability?

This is one of the major factors which is what causes 'spikes' and lagging. Loss of UDP packages reaching your PC.

 

And when they finally do arrive - the enemy is in another place.

Now you could start changing UDP packages with TCP. But then you might realize that the latency in a dogfight would be intolerable due to the increased headers and extra checks with this kind of change. The bandit would be in another place by the time you press the trigger.

TCP is not the answer, granted the time stamp is useful, but not necessary. And UDP is not the problem, it is how the data is handled that is the problem.

 

Now instead you could ask your opponent to get a better computer & router or change his ISP, so you have a better chance of killing him.

Nah, j/k - but you get my point.

I get your point, and my point is I have seen this what i described on all different sorts of servers, dedicated with many players and single PC hosted with just a few players, each with solid pings.

 

2mbps up/downlink. FC / DCS. 14-16 players in a rich environment, 500-1000 units. Good precision, very little lag / loss of UDP.

I can't see how this is bad.

 

I've flown dogfights with pilots from the other side of the world having 400-600 ms ping. Still great precision.

 

I can't see how to improve on this.

Because the latency of the data is NOT a problem!

As long as the latency is consistent.. It is when the ping times vary that you can get the problems I described above. Granted there is a limit to the size of the latency, but you get my point

 

UDP packages will get lost. You can't change this fact unfortunately.

Very true

 

But as noted above, the problem is not so much with the 'real' object pos, vel, acc data contain in the actual UDP data packet as much as the estimated (predicted aka interpolated) pos, vel, acc calculations between data packets.

 

That is where I suspect the problem lies in that the DCS netcode calculations even has trouble with planes that are not moving and just sitting on the runway wrt warping

[c0ff]

That is where I suspect the problem lies in that the DCS netcode calculations even has trouble with planes that are not moving and just sitting on the runway wrt warping

 

The planes which are 'not moving and just sitting on the runway' are actually moving slightly.

 

And the problem with netcode is temporal aliasing in the input data.

Predictor, which may be described as a digital filter can perfectly recontruct the objects trajectory, as long as the frequency content of the input data does not exceed Nyquist limit of the predictor.

To meet this requirement the sending side should filter the data before sending.

This is another can of worms and a layer of complex code ... and more latency.

That's why it is not done yet.

[shabo]

Wouldnt a dedicated server version of the sim help this?

 

Forgive my ignorance ;)

[sobek]

Wouldnt a dedicated server version of the sim help this?

 

A dedicated server can not alter the laws of signal processing theory. :)

 

It's like saying: "Wouldn't letting the sun shine at night help against all that darkness?" ;)