Once our client application has determined the address of the host machine it wants to connect to, it must determine which transport protocol to use. This section describes the Bluetooth transport protocols closest in nature to the most commonly used Internet protocols. Consideration is also given to how the programmer might choose which protocol to use based on the application requirements.
Both Bluetooth and Internet programming involve using numerous different transport protocols, some of which are stacked on top of others. In TCP/IP, many applications use either TCP or UDP, both of which rely on IP as an underlying transport. TCP provides a connection-oriented method of reliably sending data in streams, and UDP provides a thin wrapper around IP that unreliably sends individual datagrams of fixed maximum length. There are also protocols like RTP for applications such as voice and video communications that have strict timing and latency requirements.
While Bluetooth does not have exactly equivalent protocols, it does provide protocols which can often be used in the same contexts as some of the Internet protocols.
The RFCOMM protocol provides roughly the same service and reliability guarantees as TCP. Although the specification explicitly states that it was designed to emulate RS-232 serial ports (to make it easier for manufacturers to add Bluetooth capabilities to their existing serial port devices), it is quite simple to use it in many of the same scenarios as TCP.
In general, applications that use TCP are concerned with having a point-to-point connection over which they can reliably exchange streams of data. If a portion of that data cannot be delivered within a fixed time limit, then the connection is terminated and an error is delivered. Along with its various serial port emulation properties that, for the most part, do not concern network programmers, RFCOMM provides the same major attributes of TCP.
The biggest difference between TCP and RFCOMM from a network programmer's perspective is the choice of port number. Whereas TCP supports up to 65535 open ports on a single machine, RFCOMM only allows for 30. This has a significant impact on how to choose port numbers for server applications, and is discussed shortly.
UDP is often used in situations where reliable delivery of every packet is not crucial, and sometimes to avoid the additional overhead incurred by TCP. Specifically, UDP is chosen for its best-effort, simple datagram semantics. These are the same criteria that L2CAP satisfies as a communications protocol.
L2CAP, by default, provides a connection-oriented  protocol that reliably sends individual datagrams of fixed maximum length . Being fairly customizable, L2CAP can be configured for varying levels of reliability. To provide this service, the transport protocol that L2CAP is built on  employs an transmit/acknowledgement scheme, where unacknowledged packets are retransmitted. There are three policies an application can use:
retransmit until total connection failure (the default)
drop a packet and move on to queued data if a packet hasn't been acknowledged after a specified time limit (0-1279 milliseconds). This is useful when data must be transmitted in a timely manner.
Although Bluetooth does allow the application to use best-effort communication instead of reliable communication, several caveats are in order. The reason for this is that adjusting the delivery semantics for a single L2CAP connection to another device affects all L2CAP connections to that device. If a program adjusts the delivery semantics for an L2CAP connection to another device, it should take care to ensure that there are no other L2CAP connections to that device. Additionally, since RFCOMM uses L2CAP as a transport, all RFCOMM connections to that device are also affected. While this is not a problem if only one Bluetooth connection to that device is expected, it is possible to adversely affect other Bluetooth applications that also have open connections.
The limitations on relaxing the delivery semantics for L2CAP aside, it serves as a suitable transport protocol when the application doesn't need the overhead and streams-based nature of RFCOMM, and can be used in many of the same situations that UDP is used in.
Given this suite of protocols and different ways of having one device communicate with another, an application developer is faced with the choice of choosing which one to use. In doing so, we will typically consider the delivery reliability required and the manner in which the data is to be sent. As shown above and illustrated in Table 2-1, we will usually choose RFCOMM in situations where we would choose TCP, and L2CAP when we would choose UDP.
Table 2-1. A comparison of the requirements that would lead us to choose certain protocols. Best-effort streams communication is not shown because it reduces to best-effort datagram communication.
|Reliable, datagram||TCP||RFCOMM or L2CAP with infinite retransmit|
|Best-effort, datagram||UDP||L2CAP (0-1279 ms retransmit)|
The L2CAP specification actually allows for both connectionless and connection-based channels, but connectionless channels are rarely used in practice. Since sending ``connectionless" data to a device requires joining its piconet, a time consuming process that is merely establishing a connection at a lower level, connectionless L2CAP channels afford no advantages over connection-oriented channels.
The default maximum length is 672 bytes, but this can be negotiated up to 65535 bytes
Asynchronous Connection-Less logical transport