CUDA

but i often start a 5v5 which may go 40 minutes, and I die after 20 minutes
if the game is not exciting, do I want to watch 20 minutes?

It will be interesting how it works out. Maybe uber overestimates the willingness of people to pay a monthly 20€ for standard servers or maybe even a lot more for big servers, which may be needed forbig games, for an RTS. (On the other hand, really a lot of people pay monthly fees for MMORPG and such, or spend incredible amounts on vanity items on other games, or buy world of tanks premium....)
But maybe we still dont totally understand the concept of how a "standard" 1v1 to 10v10 will work in the final release.

I can understand the advantages of having a client/server model instead of P2P, but as bandwidth is more a problem than CPU these days, I'm not sure it's the right solution for a RTS. Hell, even FPS are switching back to p2p !

Also Neutrino said that he expect servers to run on lower CPUs than gaming rigs (and he is right), so it's even less a problem CPU-wise, the argument "nobody will make your game lag because of a low CPU" is kind of defeated as everybody will probably have an higher CPU than the server.

Even if Uber is hosting his array of servers, we will have to expect waiting lists (unless they get A LOT of them, but Uber is not blizzard or EA), and they need to deploy servers all around the world too. What if they stop hosting them for some reason ?

An hop in/hop out feature will really be that useful ? Is it really impossible with P2P ? (a FA mod almost does it ?)
Sure jumping in time in replays could be nice, but I don't see why it couldn't be done with a P2P protocol ? (Starcraft 2 is using routed p2p and it's allowing that).
I mean the replay file must contains everything needed to do that, the fact that it's only on the server or on each client doesn't change anything ?

On the contrary, with a server/client, it means that only the server has the replay : If it doesn't share it, you will never be able to watch any of your replay !

Couldn't have put it better myself...

P2P uses more bandwidth than Client server. Clients only need to upload/download from one machine (the server), whereas P2P requires doing so to each other computer. The worst is the Server, but even that is at the worst case equal to the bandwidth of a computer in a P2P config for the same amount of players.

It's easier to find 1 Computer that can handle the game than X computers, where X is the number of players. Also remember that the server component does not have the CPU overhead of rendering and graphics. I'd rather have the "problem" of finding 1 fast computer than having to have all the players have roughly equivalent computers.

Um, you host your own game instead? When you play single player, you are actually running the Client + Server on your own computer. You can do the same with a multiplayer game. The idea of the servers seems to be more geared towards 40 player games, or people who want an absolutely reliable connection, Not your average 1v1 or 4v4.

There's been no official word on how this will be implemented, so I won't speculate. It's certainly more difficult to add a player to a P2P game than add a client to a server, since in P2P it would have to sync up with every other player.

If it's an option, you'll no doubt see whether replays are shared before you join the game.

server-client certainly has disadvantages, but p2p isnt exactly a perfect solution, either.
I know people who stopped playing FA because of their crappy routers making them end up with 50% NIL-connections. I tried to play together with a group of friends online behind the same router a few times and of those maybe half failed because the game could not handle the connections to the different players behind the same router. Was probably the fault of the router, but is still less likely to happen if you only connect to a single server.

The slow-down problem certainly is something that p2p is suffering more from than server-client. A 40 player game would probably impossible with p2p. It is hard to find 12 people who can play a 6v6 in FA together, 40 would probably be impossible with p2p.
So server-client make bigger games possible since you only need one big server and not 40 big clients.

Since we can host our own servers normally sized games wont be that different between p2p and client-server, apart from the server-guy being able to cheat more in theory. But it's not like FA's p2p was free of cheating (i.e. there were those cheat-ui-mods that gave away a lot of pretty crucial information).
Sure the person running the server will have to let the server run until the game ends, so he wont be able to host another game as soon as he dies. I am hoping that he will be able to quit his PA-client, since the server should be an additional piece of software. So he can restart the client and join somebody else. But sure, that's a disadvantage that you have to be aware of if you are hosting. But I doubt that it will be that hard to find people willing to deal with it.

I'd guess it is not that much of a problem to send this data at the end of the game, replays are usually not that big. Sure there might be problems if the server and the client lose connection before that happens. In that case it would be best to have a global replay-repository that the replay will be uploaded to, so the clients can access it afterwards or some similar solution. Either UBER already has an idea for it, or we will make them implement something at some point during alpha/beta

In server-client the clients send commands to the server BUT the server has to send the game's state to the clients. This can certainly be optimized quite a bit, but it will probably still need more upload per client on the server.

Nope. In FA, only the commands (move there, attack that) are sent.
On a server/client protocol like uber wants to make, not only the commands are sent, but also the sim and unit state. I can't see how it can be less data, even if it only send the maths for trajectories, an origin, an initial vector and some results from time to time.

It will not if bandwidth is the problem.

That's nothing. Again, look at FA, most (if not all) of the CPU is consumed by computing the sim. It's even more true these days with modern engine/GPU and multi-threading.

Except that the dedicated server is probably less powerful than any of gaming rig from 3 years ago.
So if uber, as stated, are hoping for lot of dedicated servers, you can be sure that the CPU requirements won't be very high.

Again, problems : Bandwidth, cheats, closing the server, having to wait to play again (because of CPU and bandwidth requirement won't make you able to run lot of games concurrently).
ie. you play a big game on the evening, you die first, you want to go to sleep and shut down the PC : You can't.

Yes, but my question is : Is is actually needed ? Do you really want to join a game in a middle of a battle ? We are talking of a RTS, not a FPS.

It's true, but it's avoidable. Call of Duty is using peer to peer, starcraft 2 is using routed peer to peer, no nil problems AFAIK.

By the way, I'm still searching a tech guy to implement a router peer to peer function in FAF for players having nil.

I can remember having problems to connect in some Call of Duty, it wasn't permant, though.
Starcraft 2 certainly is very very good at connecting people together, but if UBER would declare that they will route ANY game traffic through their own server people would probably accuse them of breaking their no-drm-promise. Sure they could make that routing-server a public piece of software but that would bring us back to a server-client-concept, but one that loses the advantage of the server doing the heavy work for the clients.

You only need a single person who has a really good connection and computer. Currently for a 4v4 you need 8 people with a decent connection and a good computer. PA aims to make it possible to play way bigger games, so p2p really is not an option.
I don't know any statistical data on it, but if you take a random group of 10 people out of the PA-community some of the will probably have the connections required.
Bandwidth was a HUGE problem for FA with it's p2p. There were and still are people who cannot play 4v4, because their upload cannot handle it. Back in 2008 my own connection was barely able to deal with 4v4. These same people will probably be able to participate in 40 player games in PA as client. I'd call that an improvement.

Just because many servers wont be really powerful doesn't change anything about raeven's argument. It still is easier to find one powerful server and 7 weak clients to play a 4v4 than to find 8 powerful clients.

I don't understand what that is good for either. ^^

I really don't see that as such a terribly big problem. If you open a server you just need to know that you will be responsible of running it until the game is over. If you don't want that don't open a server.


We're going in circles with this discussion. p2p and server client both have advantages and disadvantages, the guys at UBER probably did think this whole discussion through and came to the conclusion that server client is the better choice for PA. I doubt anything that we are writing now can even change it anymore, since from what they wrote it seems that their server-client implementation is already halfway through development.

If this was even remotely true, FA wouldn't demand a constant 24kB per player. The bandwidth requirements for large player games was pretty retarded.

it is true. what else should be send? You can see that this is true, when a desync happens Then you will see that all games have totally different outcomes, as the commands are trensferred between the different players only partially or not at all.

There is no possible way you could make the game I'm envisions as P2P. Not possible.

If you want to limit it to 8 players and a smaller sim then sure. I don't think people are fully processing the "scale" thing I keep talking about yet Why? People have a hard time envisioning things that haven't been done yet.

The bandwidth thing is complicated. At the 4-8 player size I would argue we will be using more bandwidth. Once you start having games with more players the amount of bandwidth that a client uses remains more or less steady per player as we are only sending the slice of the world they can see at any point. So at this point the server based setup starts to win.

It also wins at NAT traversal, leave/rejoin, impossibility of having a desync (how big a game can you really have if you eventually desync when the sim gets big), ability to multithread the sim properly (multithreading a sync sim is very very difficult), replays that are robust across versions (and lots more replay awesomeness) and more.

Also keep in mind that although TA used a p2p network model the code itself was closer to how PA does it in terms of what gets sent over the network. For a large scale game sync sim is a disaster.
Using sync on SupCom was a mistake in retrospect.

Pardon if this is a silly question, but would it be at all possible to enable a P2P mode? I only ask because I prefer 1v1s in strategy games (I play team, but not much) and will probably play PA for 5-10 years. Will there be any way I can play it without needing an external server? I hope the answer is yes because I consider this to be critical.

You could find a server made for 1v1.
You could just run your own server on your PC or an old PC.
You could play LAN matches with your friends.

You have many options.

It's not necessary. You can host the game(flip a coin or pick the stronger computer), and there's a pretty good chance that 1v1 servers will exist.

It might be possible to distribute the hosting burden among multiple separate PCs (which is something P2P doesn't help with anyway). The bandwidth needs would be too demanding for online, but it may succeed on a local network. I won't claim how easy or difficult that might be.

Where is the big difference in case of 1v1 if one of the players hosts the game?

So run your own server. We aren't restricting you to playing on hosted servers.

Does this mean we can host our own PA server and or local game if we want to? so that we can play with our friends in the cabin in the woods where we have no internet?

let the strongest pc run the server and you get a single player +x scenario.
How well this works depends a bit on the size of x.
But if you have enough friends who are willing to travel to a cabin in the woods to play computer games, the probability, that one of them has a half-way decent spare computer around to run a dedicated server, isn't that low.

This would be good that the best computer should run it for lan and offline play

What about a peer to peer with a star topology ?
Each client send their command to a single server/player, and that single player/server send the command of everyone to everyone.

As you state that you think dedicated server will be smaller machine than gaming PC (I agree on that), computing the sim on every PC would allow bigger games more easily.

And that kind of networking will normally run with a lower bandwidth requirement for everyone (server and client) than a full client/server solution.

That also allow the central server to drop and another client to take over, as each client will be aware of the full sim, unlike the solution you are currently aiming to.

CPU is less an issue than bandwidth and server costs these days, did you think about it and denied that idea, if so, why ?