[SDL2 – Part 13] Multiplayer – TCP

Continuing where we left of…


Last time we left off after having taken a look at UDP connections. But, as we saw, UDP connections have several issues :

  • No delivery guarantee
  • Packages might not arrive in the same order they were sent

So how do we solve these issues? By creating an actual connection. A TCP connection that is.

TCP connections


As mentioned in the previous post, UDP connections aren’t connections at all. It’s basically just two parts sending data back and forth between them. TCP connections, however, are proper connections. This means we know the following :

  • Both parts tries to keep the connection alive
  • Both parts have received everything the other part has sent
  • Every package has been received in order

These points requires two different mechanisms, so let’s look at the one by one.

How TCP works


In order to have a connection, both parts have to agree on it. This is different from UDP connection because in UDP connections we just send the data without bothering about actual connections. So how do we set up the TCP connection? By using a technique called three-way handshake

The three-way handshake


The process of establishing a TCP connections requires three different steps:

  • Part 1 contacts part 2 and asks to start a connection
  • Part 2 replies and either says OK, or declines
    • If part 2 declines, the process stops here
  • Part 1 replies back, to confirm that he’s received part 2’s request
  • Now the actual connection is started and we can send data

The third step might seem kinda unnecessary, but remember that any packages can get lost. And so if part 2 doesn’t get the answer from part 1, part 2 doesn’t know whether part 1 is still there. Part 1 might have lost the connection or wished to stop it altogether. We can compare this to starting a phone call :

  1. Part 1 calls part 2
  2. Part 2 picks up and says hello
  3. Part 1 says hello back

After the last step, the conversation is started. All of these step are necessary for us to know there is an actual conversation. Skipping one of them might make you think something is wrong. For instance, if part 2 doesn’t say ‘hello’, you might think there is something wrong with the connection.

Dealing with the UDP issues


The major flaw of UDP is that it doesn’t deal with loss of packages. And another issue is that we don’t know if we’re getting the correct package at the correct time. Potentially we can get every package in the wrong order. Which could lead to everything from glitchy gameplay to crashes. So let’s look at the mechanics that makes TCP connections able to give their guarantees :

ACK numbers


In TCP every package gets a unique number. This is basically just a counter so the first package gets the number 0, next one 1 and so on…

This number is then sent back from the recipient to confirm that “okay, so now I have every package up to this package” ( I will refer to the confirmation as ACK numbers.) So when we get an ACK for the last paackage we sent, we know that the other part has received everything we have sent so far. This means we don’t have to worry about any of the packages having been lost.

But say the receiver misses a package in the middle? For example, what if the receiver gets package 1,2,4, but not 3? The receiver will look at the ACK numbers and think “huh… I’m missing a package here.” And will only send ACK 2 back. At this point, your application might get package 1 and 2, but not 4, since it’s out of order.

Flow of TCP


Let’s look at an example to see how the TCP might handle loss of packages.

  1. Sender sends package 1,2,3,4
  2. Receiver receives package 1
    • Your application receives package 1
    • Send ACK 1 all good so far!
  3. Receiver receives package 2 and 4
    • Uh oh! Where’s package 3?
  4. Sender receives ACK 2
  5. Sender waits a while before resending package 3 and 4
    • It sends package 4, which the receiver already has, because it only got ACK2
  6. Receiver receives package 3
    • Now your application will get the last two packages ( 3 and 4 )
    • Receiver sends ACK 4
  7. Sender receives ACK 4
  8. Receiver receives package 3, again!
    • The original package 3 was just delayed get it again.
    • Since we already have these packages, it’ll get discarded

Now the receiver has got all 4 packages and all is fine and dandy. But if you study the example, you see that in our case the sender has to resend 2 packages ( 3 and 4), while the receiver in reality only needed one ( 3 ) And in addition, we got package 3 twice. But even though we got package 4 early, it wasn’t sent to our application before we got package 3 because TCP we receive everything in order. This is one of the major drawback of TCP ; there can be lots of delays and re-sending of packages.

TCP send rate


A final issue about TCP performance is how it regulates package sending. TCP can send several packages at any time. The basic idea is that every time every package is sent successfully and in time, it’ll send more packages the next time.

So the first time, it might only send 2 packages the first time. But if it gets ACK for all those three in a given time, it might send 4 the next time. Then maybe 8 and keep increasing it until it doesn’t get an ACK for all packages on time. When that happens, it’ll send less packages the next time. Let’s look at a simple example :

  1. Send 2 packages
    • Receive ACK for both packages
  2. Send 4 packages
    • Receive ACK for all packages
  3. Send 8 packages
    • Receive ACK for only 5 packages
    • We’re missing 3 packages! Maybe this was too many packages? Try sending less…
  4. Send 6 packages

As you can see, TCP will try its best to keep sending the maximum number of packages without having to resend anything.

Complexity of TCP


Although TCP is quite old ( more than 30 years old, ) it’s really complicated. There are a lot of different mechanisms involved to deal with anything that might happen. Not only do they handle the cases we’ve seen, but it also, as we saw, needs to control the speed for optimal performance.

I have purposely simplified it because its nice to have a basic understanding of how TCP works as this might help you to choose whether you should use UDP or TCP

The two different parts of TCP connections


Since TCP connection are actual connections, there needs to be a central part ( server ) that the others ( clients ) connect to. I’ll briefly discuss servers and clients, and what their roles are when it comes to TCP

Server


The server is the part that all the clients will connect to. So the server will always be listening for new connection and accepting them as they come. The server will be accepting connections from any client, so we don’t specify IPs on their server side ( ,ore on this later. ) We will only specify the port to listen to.

Client


The client tries to connect to the server. It needs to know both the IP and port of the server before it can try to connect to the server. The server doesn’t know anything about the client until it tries to connect to the server.

When the server accepts the client, the connection is established. At this point in time, the server also has a connection to that client specifically. We’ll see this later.

Note that these are just different types of connection, not necessarily different computers. A computer can have any number of server and / or client connections.

Time for some code


So now that we have all the technical information about TCP is out of the way, we can start setting up a TCP connection of our own. As you might expect, the source code for TCP is a bit more involved than the UDP one.

This part relies on code from my previous post. If something is unclear, you can go back and read that part if you want more information. I have also added links to the documentation in the headers.

SDLNet_ResolveHost


Just like with UDP, we need to ask SDL net to get correct the correct representation ( because of different endianness). The function is the same ( but it is used in a different way for servers, so do read on)

Parameters :

  • IPaddress* address – a pointer to an allocated / created IPAdress.
  • const char* hostIP address to send to ( xxx.xxx.xxx.xxx )
  • Uint16 port – the port number to send to

Return value

The int value 0 on success, otherwise -1. In this case, address.host will be INADDR_NONE. This can happen if the address is invalid or leads to nowhere.

But there is a slight difference in how it’s used. Since TCP connections are actual connections it has both a server and a client part :

Server

The server part needs to be listening for IPs that are trying to connect. So we’re not really resolving a host this time. We’re just preparing the IPaddress for the next step.

So what we do is that we simply use null as the IP. This tells SDL_net that we don’t want to use this IPaddress to connect to a host, but rather just listen for other connections. SDL_net does this by setting the IP to INADDR_NONE. This comes into play in the next step ( SDLNet_TCP_Open)

Client

For clients, this function is used more or less exactly like in the previous part ; it’ll prepare the IPaddress with the information we supply.

Of course, the port on both the server and client side has to be the same.

Note: no connections has been initiated yet, we’ve just asked SDL to prepare the port and IP address for us.

TCPSocket


This is a new type. It represents a TCP connection. We’ll use it just like UDPSocket but of course this time it represents a TCP connection instead of a UDP connection.

SDLNet_TCP_Open


Now that we have an IPadress correctly set up, we can try to connect to a host. This is done by the function SDLNet_TCP_Open

Here is the function signature.

Parameters :

  • IPaddress *ip – an IPaddress* contaning the IP and port we want to connect to. We’ll use the one we got from SDLNet_ResolveHost

Return:

  • For clients : a TCPsocket to the server, which can be used for sending and receiving data
  • For servers : a TCPsocket used for listening for new clients trying to connect

This function will try to open a connection. But just like with SDLNet_ResolveHost, there are two different cases here

Server


Above we saw that if we call SDLNet_ResolveHost with null as the IP, SDL_net will set the IP of the to INADDR_NONE. This means we will be listening for connections, rather than trying to connect. This is because, as a server, we don’t actively try to connect to another host ( we just accept connections ), so we don’t know about any IP address yet.

What this function does in this case, is that it tries to open the port for listening.

Client


For clients, this works much like for UDP : we try to connect to the server with the given IP and port

At this point, the client is connected to the server, and now they can communicated. This is a little different from how it works in UDP so let’s start by looking at how the communcation can be done in TCP

A quick example


Before we jump into the next part, let’s have a quick look at an example of how to use these two functions. These two functions are the initialization part of the TCP code. Since these steps are slightly different form client to server, I’ll cover them separately.

Server


Simply set up the IP address and use it to open a port for listening :

Client


Simply set up the IP address and try to connect to server with that IPaddress :

The job of the client


The clients are the parts you’ll be dealing with they most. A client communicates with other clients. This is more or less just like in UDP, but there are som differences.

SDLNet_TCP_Send


Sending data using TCP is done using a slightly different function from that of UDP :

Parameters :

  • TCPsocket sock – the TCPsocket* on which to send the data
  • const void *data – the data to send
  • int len – the length of the data ( in bytes )

This function is quite straight ahead. The only thing to note is the void*. The type void* is something that is widely used in C but not so much in C++. It’s basically a pointer to anything. So the data can be just about any form of data. This requires a bit of low-level C “hacking” to get right.

Return:

The number of bytes that was sent. If this is less than the size of the data we tried to send ( or the len parameter, ) an error has occured. This error could be the client disconnecting or a network error.

Using this function correctly is tricky, in a similar way to UDP. Let’s look at a possible way to implement it ;

SDLNet_TCP_Recv


Receiving data using TCP is also done using a slightly different function from that of UDP :

Parameters :

  • TCPsocket sock – the TCPsocket* on which to recv the data from
  • const void *data – the data to receive
  • int maxlen – the maximum data to receive

Return:

The number of data received. If this is less than 0, an error has occured.

And since this is C ( and not C++ ) we need to allocate a decent sized buffer in advance ( this is the void *date part. It’ll have the same size as maxlen. The setting of the buffer involves a little C-style trickery.

Let’s look at an example :

The job of the server


So now we have a TCPsocket that listens to the port we specified. And now we can try to accept new connections. For now, we’ll try to accept connections right out of the blue. But later we’ll look out how to check for clients trying to connect. Anyways ; here is the method we need:

SDLNet_TCP_Accept


This is the essentially the accept part of the three-way handshake. The client has tried to connect to us and we need to accept it before the connection is established. This function does exactly what you might expect : it accepts an incoming TCP connection, informs the client and thus establishing the connection.

Parameters :

  • TCPsocket *server – the TCPsocket* we use for listening for new connections. This is the TCPConnection we created using SDLNet_TCP_Open.

Return :

A different TCPsocket this TCPsocket does represent a connection to a specific client. If it’s valid, it means a connection has been established. If it’s null it means no connection was established. This can mean that there was an error. But it can also mean that there was no clients trying to connect.

This function might lead to some confusion as there are two TCPsockets, but remember :

The first one ( the parameter we supply ) is ther server TCPsocket. This is not connected to any client, we just need it to be able to listen for new connection. We create this TCPSocket by callling SDLNet_TCP_Open

The second TCPsocket is for a specific client.We create this TCPSocket by callling SDLNet_TCP_Accept. When it’s created, it can be used exaclty like the TCPsockets created on the client side. ( As I talked about in the cleint part of SDLNet_TCP_Open )

Dealing with SDLNet_TCP_Recv


There is a major issue with the receive function. It blocks. This means the function waits until it has received something. Actually, according to the documentation, it’ll wait til it has received exactly maxlen bytes and then set those in the void* data. But from what I’ve found, this isn’t 100% true.

What I have found, is that the function will block. But only until it has received something ( at most maxlen bytes. ) So, in other words, it waits til it has received something, no matter how little or much it is. But even though this is better than waiting for maxlen bytes, the fact that it blocks is still an issue we’ll need to solve.

SDLNet_TCP_Recv will also join together messages if it can. So say client 1 sends

“Hello”

and

“World”

in two separate messages, SDLnet can join them together so that what client 2 gets is

“HelloWorld”

in one message.

This can ( and probably will ) happen if buffer size is large enough.

Or, if the buffer size is too small one call might only get part of the data. So if client 1 sends :

“HelloWorld”

But client 2 has the buffer size set to 6, it’ll get

“HelloW”

The first time client 2 calls SDLNet_TCP_Recv. And

“orld”

The second time it calls SDLNet_TCP_Recv

That means there are two issues to fix : the fact that it blocks and the fact that we might not receive everything with one call to SDLNet_TCP_Recv.

SDLNet_SocketSet


To solve this, we can check if something has happened on a collection of TCPsockets, this includes someone connecting, disconnecting or receiving data.

We can use a SDLNet_SocketSet to solve this. Think of it as simply a set of sockets. We’ll be using it for storing and checking TCPsockets to see if there is any activity. A SDLNet_SocketSet can contain any number of TCPSockets. Those can be both server and client connections.

SDLNet_TCP_AddSocket


This is a really simple function for adding a socket to a SDLNet_SocketSet. It also exists for UDP, but we’ll be using the TCP version, of course.

Parameters :

  • SDLNet_SocketSet *set – the SDLNet_SocketSet we want to add the TCPsocket to
  • TCPsocket *sock – the TCPsocket we want to add to the SDLNet_SocketSet

Return :

The number of TCPsockets in the SDLNet_SocketSet on success. Or -1 on failure.

SDLNet_CheckSockets


Now that we’ve added sockets to the SDLNet_SocketSet, we can use the SDLNet_CheckSockets function to check for activity. “Activity” in this case basically means that something has happened. This can either mean we have received data, that someone has disconnected or that there is an error.

Parameters :

  • SDLNet_SocketSet *set – the SDLNet_SocketSet we want to check for activity
  • Uint32 timeout – a variable stating how long ( in milliseconds ) we want to wait for activity. We can wait anything between 0 milliseconds and… well anything up to 49 days.

Return :

The number of TCPsockets in the SDLNet_SocketSet with activity on success. Or -1 if either theSDLNet_SocketSet is empty or there was an error.

SDLNet_SocketReady


After we’ve called SDLNet_CheckSockets, we can use this function to check whether a particular TCPSocket has been marked as active. This function should be called on a socket on a SDLNet_SocketSet after code>SDLNet_CheckSockets has been called on the SocketSet that holds that TCPSocket.

Parameters :

  • TCPSocket *socket – the TCPSocket we want to check for activity

Return :

Count of TCPSockets with activity

In other words ; we use SDLNet_CheckSockets to see if any of the TCPSockets in a SDLNet_SocketSet has any activity. If so, we can call SDLNet_SocketReady on each of the SDLNet_SocketSets in that SDLNet_SocketSet to see if that TCPSocket in particular has any activity.

Examples

Now let’s look at how you could implement an update function that checks for activity. They’ll be different for server and client connections since client connections checks for incoming messages and disconnections. While on the server side we’ll simply check for clients trying to connect.

Client side example


As I mentioned above, on the client side we need to check for connections and incomming messages. Here is a way to do that :

A problem that arises here, is that calling code>SDLNet_CheckSockets kind of sets the TCPSocket back to “inactive” when you call it. Even if there is several messages waiting to be read.

So when you have called ReadMessage(), you have no way of knowing if it has any more data. Calling it again, would mean calling SDLNet_TCP_Recv again which could block until the other client sent more data.

This is an issue lots of tutorials that I’ve seen has. But there is a solution that doesn’t block ; we just need to call SDLNet_CheckSockets again. So just add this to the bottom of the previous function

Server side example


On the server side, we need to check for clients trying to connect. This is fortunately a little bit simpler than what we had to do on the client side. Here is the code :

I think that’s all for now. You can find a working implementation

here.

Conclusion


Setting up a TCP connection using SDL_Net is quite tricky. Lots of tutorials out there just briefly discuss the topic without going into much detail about the different functions. Hopefully this post has helped you get a better view of the different parts of SDL_net ( I sure did writing it! ) I might also post a third networking post about an even better way of doing network communication using both UDP and TCP for maximum performance.

I’m also really glad to finally have finished and published a new post. I know it’s been a long time since last time, but I’ve been a bit busy at work and haven’t really had the time or energy. But I feel some of my energy is back. And getting positive feedback is always amazing, they help me keep going. So thanks to everyone who’s commented! : )

(Semi) Final code :

Working implementation of TCP connections ( NOTE : work in progress! )

Github page


Feel free to comment if you have anything to say or ask questions if anything is unclear. I always appreciate getting comments.

You can also email me : olevegard@headerphile.com