Connection-oriented service
The service a protocol provides to its consumers is connection oriented when each party involved negotiates a virtual connection before sending the actual data. During the setup process, a number of parameters about the connection must be agreed upon. This is analogous to the older wired telephone systems, where a dedicated connection is set up between the two hosts. In modern networks, an example is TCP. The PDU for TCP is a segment, which consists of a header and a data section. The header has a few fields which are used to transition between states of the protocol state machine. The next figure shows what the TCP header looks like in practice. Each of the rows in this figure are of 32 bits (thus, each row is two octets), and some are divided into multiple segments:
We will look at a few of these which are used for manipulating the connection between hosts:
- Control bits (flags) are a set of 9 bits that are used for various purposes. The flags of interest here are SYN, ACK, FIN, and RST. SYN triggers a synchronization of sequence numbers. The ACK flag indicates that the receiver should care about the corresponding acknowledgment number. The FIN flag starts the process of tearing down a connection. The RST flag resets the connection in case of an error.
- The sequence number is a 32-bit field which is used to reorder messages at the receiver. When the SYN flag is set (which should be the case only for the first packet in a connection), the sequence number is the initial sequence number; otherwise, it is the sequence number accumulated so far.
- The acknowledgement number is a 32-bit field which is used to enable reliable delivery of messages. If the ACK flag is set, this value is the next sequence number that the sender is expecting.
Before two hosts running TCP can start exchanging data, they must do a three-way handshake to establish a connection. This works like this: the client that wants to initiate communication sends a SYN packet to the server. The sequence number is set to a random value and the SYN flag is set to 1. The server responds with a packet that has both SYN and ACK set to 1. This packet has the acknowledgment number set to one more than what it got from the client, and the sequence number is set to a random number. Finally, the client responds with a packet that has the ACK flag set, the sequence number set to the received acknowledgement number in the last step, and the acknowledgement number is set to one more than the sequence number in the previous step. After this is done successfully, both the client and the server have agreed on sequence and acknowledgement numbers. The advantage of this model is that is has a reliable connection where both the sender and the receiver knows what to expect. The sender can tune the rate of sending data, depending on how fast or slow the receiver is or how congested the network is. The disadvantage here is the higher connection setup costs. Assuming it takes 100 ms to send a packet to a host in another continent, we will need to exchange at least 3 packets before we can begin sending data. That amounts to a delay of 300 ms. While this might not look like a lot, remember that at any given point, a laptop being used for accessing Facebook might have thousands of connections open to servers all over the world. The connection oriented service model works fine for a large number of use cases, but there are a few cases where the overhead is either significant or unnecessary. An example is video streaming. In this case, a few missing packets do not cause a huge problem since no one notices a small number of misaligned pixels in the video. These applications prefer a connectionless model, as described below.