Jenkins Software

Creating Packets
How to encode your game data into packets

Systems that run RakNet, and in fact all systems on the internet, communicate through what is commonly known as packets. Or more accurately in the case of UDP, datagrams. Each datagram is created by RakNet and contains one or more messages. Messages could be created by you, such as position or health, or sometimes are created by RakNet internally, such as pings. By convention, the first byte of messages contain a numerical identifier from 0 to 255 which is used to indicate what type of message it is. RakNet already has a large set of messages it uses internally, or for plugins. These can be viewed in the file MessageIdentifiers.h.

For this example, lets set the position of a timed mine in the gameworld. We'll need the following data:
  • The position of the mine, which is 3 floats. float x, float y, float z. You may have your own vector type which you can use intead.
  • Some way to refer to the mine that all systems agree on. The NetworkIDObject class is perfect for that. Lets assume class Mine inherits from NetworkIDObject. Then all we have to store is get the NetworkID of the mine (for more information see Receiving Packets, Sending Packets.
  • Who owns the mine. That way if someone steps on it we know who to give credit to. The built in reference to players, SystemAddress, is perfect.. You can use GetExternalID() to get the SystemAddress.
  • When the mine was placed. Lets say after 10 seconds the mine is automatically disintegrated, so it's important that we get the time correct so the mine doesn't disintegrate at different times on different computers. Fortunately the RakNet has the built-in ability to handle this using Timestamping.
Use a structure or a bitstream?

Ultimately, anytime you send data you will send a stream of characters. There are two easy ways to encode your data into this. One is to create a structure and cast it to a (char*) the other is to use the built-in BitStream class.

The advantage of creating a structure and casting is that it is very easy to change the structure and to see what data you are actually sending. Since both the sender and the recipient can share the same source file defining the structure, you avoid casting mistakes. There is also no risk of getting the data out of order, or using the wrong types. The disadvantage of creating a structure is that you often have to change and recompile many files to do so. You lose the compression you can automatically perform with the bitstream class. And RakNet cannot automatically endian-swap the structure members.

The advantage of using a bitstream is that you don't have to change any external files to use it. Simply create the bitstream, write the data you want in whatever order you want, and send it. You can use the 'compressed' version of the read and write methods to write using fewer bits and it will write bools using only one bit. You can write data out dynamically, writing certain values if certain conditions are true or false. BitStream will automatically endian-swap members written with Serialize(), Write(), or Read(). The disadvantage of a bitstream is you are now susceptible to make mistakes. You can read data in a way that does not complement how you wrote it - the wrong order, a wrong data type, or other mistakes.

We will cover both ways of creating packets here.
Creating Packets with structs


As I’ve probably mentioned earlier, RakNet uses a convention on how packets (Packet) types are identified. The first byte of the data field is a single byte enumeration that specifies type, followed by the transmitted data. In packets that include a time stamp, the first byte contains ID_TIMESTAMP, the following 4 bytes the actual time stamp value, then the byte that identifies the packet, and only then the actual data transmitted.

Without Timestamping

#pragma pack(push, 1)
struct structName
{
unsigned char typeId; // Your type here
// Your data here
};
#pragma pack(pop)

Noticed the #pragma pack(push,1) and #pragma pack(pop) ? These force your compiler (in this case VC++), to pack the structure as byte-aligned. Check your compiler documentation to learn more.

With Timestamping

#pragma pack(push, 1)
struct structName
{
unsigned char useTimeStamp; // Assign ID_TIMESTAMP to this
RakNet::Time timeStamp; // Put the system time in here returned by RakNet::GetTime() or some other method that returns a similar value
unsigned char typeId; // Your type here
// Your data here
};
#pragma pack(pop)

Note that when sending structures, RakNet assumes the timeStamp is in network order. You would have to use the function BitStream::EndianSwapBytes() on the timeStamp field to make this happen. To then read the timestamp on the receiving system, use if (bitStream->DoEndianSwap()) bitStream->ReverseBytes(timeStamp, sizeof(timeStamp). This step is not necessary if using BitStreams.

Fill out your packet. For our timed mine, we want the form that uses timestamping. So the end result should look like the following...

#pragma pack(push, 1)
struct structName
{
unsigned char useTimeStamp; // Assign ID_TIMESTAMP to this
RakNet::Time timeStamp; // Put the system time in here returned by RakNet::GetTime()
unsigned char typeId; // You should put here an enum you defined after the last one defined in MessageIdentifiers.h, lets say ID_SET_TIMED_MINE
float x,y,z; // Mine position
NetworkID networkId; // NetworkID of the mine, used as a common method to refer to the mine on different computers
SystemAddress systemAddress; // The SystenAddress of the player that owns the mine
};
#pragma pack(pop)

As I wrote in the comment above, we have to define enums for our own packets types, so when the data stream arrives in a Receive call we know what we are looking at. You should define your enums as starting at ID_USER_PACKET_ENUM, like this:


// Define our custom packet ID's
enum {
ID_SET_TIMED_MINE = ID_USER_PACKET_ENUM,
// More enums....
};

NOTE THAT YOU CANNOT INCLUDE POINTERS DIRECTLY OR INDIRECTLY IN THE STRUCTS.

It seems to be a fairly common mistake that people include a pointer or a class with a pointer in the struct and think that the data pointed to by the pointer will be sent over the network. This is not the case - all it would send is the pointer address

Nested Structures

There is no problem with nesting structures. Just keep in mind that the first byte is always what determines the packet type.

#pragma pack(push, 1)
struct A
{
unsigned char typeId; // ID_A
};
struct B
{
unsigned char typeId; // ID_A
};
struct C // Struct C is of type ID_A
{
A a;
B b;
}
struct D // Struct D is of type ID_B
{
B b;
A a;
}
#pragma pack(pop)

 Creating Packets with Bitstreams

Write less data with bitstreams

Lets take our mine example above and use a bitstream to write it out instead. We have all the same data as before.

MessageID useTimeStamp; // Assign this to ID_TIMESTAMP
RakNet::Time timeStamp; // Put the system time in here returned by RakNet::GetTime()
MessageID typeId; // This will be assigned to a type I've added after ID_USER_PACKET_ENUM, lets say ID_SET_TIMED_MINE
useTimeStamp = ID_TIMESTAMP;
timeStamp = RakNet::GetTime();
typeId=ID_SET_TIMED_MINE;
Bitstream myBitStream;
myBitStream.Write(useTimeStamp);
myBitStream.Write(timeStamp);
myBitStream.Write(typeId);
// Assume we have a Mine* mine object
myBitStream.Write(mine->GetPosition().x);
myBitStream.Write(mine->GetPosition().y);
myBitStream.Write(mine->GetPosition().z);
myBitStream.Write(mine->GetNetworkID()); // In the struct this is NetworkID networkId
myBitStream.Write(mine->GetOwner()); // In the struct this is SystemAddress systemAddress

If we were to send myBitStream to RakPeerInterface::Send it would be identical internally to a casted struct at this point. Now lets try some improvements. Lets assume that a good deal of the time mines are at 0,0,0 for some reason. We can then do this instead:

unsigned char useTimeStamp; // Assign this to ID_TIMESTAMP
RakNet::Time timeStamp; // Put the system time in here returned by RakNet::GetTime()
unsigned char typeId; // This will be assigned to a type I've added after ID_USER_PACKET_ENUM, lets say ID_SET_TIMED_MINE
useTimeStamp = ID_TIMESTAMP;
timeStamp = RakNet::GetTime();
typeId=ID_SET_TIMED_MINE;
Bitstream myBitStream;
myBitStream.Write(useTimeStamp);
myBitStream.Write(timeStamp);
myBitStream.Write(typeId);
// Assume we have a Mine* mine object
// If the mine is at 0,0,0, use 1 bit to represent this
if (mine->GetPosition().x==0.0f && mine->GetPosition().y==0.0f && mine->GetPosition().z==0.0f)
{
myBitStream.Write(true);
}
else
{
myBitStream.Write(false);
myBitStream.Write(mine->GetPosition().x);
myBitStream.Write(mine->GetPosition().y);
myBitStream.Write(mine->GetPosition().z);
}
myBitStream.Write(mine->GetNetworkID()); // In the struct this is NetworkID networkId
myBitStream.Write(mine->GetOwner()); // In the struct this is SystemAddress systemAddress

This can potentially save us sending 3 floats over the network, at the cost of 1 bit.

Common mistake!

When writing the first byte to a bitstream, be sure to cast it to (MessageID) or (unsigned char). If you just write the enumeration directly you will be writing a full integer (4 bytes).

Right:
bitStream->Write((MessageID)ID_SET_TIMED_MINE);

Wrong:
bitStream->Write(ID_SET_TIMED_MINE);

In the second case, RakNet will see the first byte is 0, which is reserved internally to ID_INTERNAL_PING, and you will never get it.


Writing strings

It is possible to write strings using the array overload of the BitStream. One way to do it would be to write the length, then the data such as:

void WriteStringToBitStream(char *myString, BitStream *output)
{
output->Write((unsigned short) strlen(myString));
output->Write(myString, strlen(myString);
}

Decoding is similar. However, that is not very efficient. RakNet comes with a built in StringCompressor called... stringCompressor. It is a global instance. With it, WriteStringToBitStream becomes:

void WriteStringToBitStream(char *myString, BitStream *output)
{
stringCompressor->EncodeString(myString, 256, output);
}

Not only does it encode the string, so the string can not easily be read by packet sniffers, but it compresses it as well. To decode the string you would use:

void WriteBitStreamToString(char *myString, BitStream *input)
{
stringCompressor->DecodeString(myString, 256, input);
}

The 256 in this case is the maximum number of bytes to write and read. In EncodeString, if your string is less than 256 it will write the entire string. If it is greater than 256 characters it will truncate it such that it would decode to an array with 256 characters, including the null terminator.

RakNet also has a string class, RakNet::RakString, found in RakString.h

RakNet::RakString rakString("The value is %i", myInt);

bitStream->write(rakString);

RakString is approximately 3X faster than std::string.

Use RakWString for Unicode support.

Programmer's notes:

1. You can also write structs directly into a Bitstream simply by casting it to a (char*). It will copy your structs using memcpy. As with structs, it will not dereference pointers so you should not write pointers into the bitstream.
2. If you use a string very commonly, you can use the StringTable class instead. It works the same way as StringCompressor, but can send two bytes to represent a known string.

 See Also
Index
Sending Packets
Receiving Packets
Timestamping