TCP provides a communication service at an intermediate level between an application program and the Internet Protocol (IP). That is, when an application program desires to send a large chunk of data across the Internet using IP, instead of breaking the data into IP-sized pieces and issuing a series of IP requests, the software can issue a single request to TCP and let TCP handle the IP details.
IP works by exchanging pieces of information called packets. A packet is a sequence of bytes and consists of a header followed by a body. The header describes the packet's destination and which routers on the Internet to use to pass the packet along—generally in the right direction—until it arrives at its final destination. The body contains the data which IP is transmitting. When IP is transmitting data on behalf of TCP, the contents of the IP packet body is TCP data.
Due to network congestion, traffic load balancing, or other unpredictable network behavior, IP packets can be lost or delivered out of order. TCP detects these problems, requests retransmission of lost packets, rearranges out-of-order packets, and even helps minimize network congestion to reduce the occurrence of the other problems. Once TCP at the receiving end has finally reassembled a perfect copy of the large data chunk originally transmitted, it passes that single chunk up to the application program at the receiving end. Thus, TCP greatly simplifies the application programmer's network communication task.
TCP is used extensively by many of the Internet's most popular application protocols and resulting applications, including the World Wide Web, E-mail, File Transfer Protocol, Secure Shell, and some streaming media applications.
However, because TCP is optimized for accurate delivery rather than timely delivery, TCP sometimes incurs relatively long delays (in the order of seconds) while waiting for out-of-order messages or retransmissions of lost messages, and it is not particularly suitable for real-time applications such as Voice over IP. For such applications, protocols like the Real-time Transport Protocol (RTP) running over the User Datagram Protocol (UDP) are usually recommended instead.
TCP is a reliable stream delivery service that guarantees delivery of a data stream sent from one host to another without duplication or losing data. Since packet transfer is not reliable, a technique known as positive acknowledgment with retransmission is used to guarantee reliability of packet transfers. This fundamental technique requires the receiver to respond with an acknowledgment message as it receives the data. The sender keeps a record of each packet it sends, and waits for acknowledgment before sending the next packet. The sender also keeps a timer from when the packet was sent, and retransmits a packet if the timer expires. The timer is needed in case a packet becomes lost or corrupt.
TCP (Transmission Control Protocol) consists of a set of rules: for the protocol, that are used with the Internet Protocol, and for the IP, to send data "in a form of message units" between computers over the Internet. At the same time that the IP takes care of handling the actual delivery of the data, the TCP takes care of keeping track of the individual units of data "packets" (or more accurately, "segments") that a message is divided into for efficient routing through the net. For example, when an HTML file is sent to you from a Web server, the TCP program layer of that server takes the file as a stream of bytes and divides it into segments, numbers the segments, and then forwards them individually to the IP program layer. The IP program layer then turns each TCP segment into an IP packet by adding a header which includes (among other things) the destination IP address. Even though every packet has the same destination IP address, they can get routed differently through the network. When the client program in your computer gets them, the TCP stack (implementation) reassembles the individual segments and ensures they are correctly ordered as it streams them to an application.
|Bit offset||Bits 0–3||4–7||8–15||16–31|
|0||Source port||Destination port|
|96||Data offset||Reserved||CWR||ECE||URG||ACK||PSH||RST||SYN||FIN||Window Size|
Unlike TCP's traditional counterpart, User Datagram Protocol, which can immediately start sending packets, TCP provides connections that need to be established before sending data. TCP connections have three phases:
At this point, both the client and server have received an acknowledgment of the connection.
TCP uses a sequence number to identify each byte of data. The sequence number identifies the order of the bytes sent from each computer so that the data can be transferred reliably and in order, regardless of any fragmentation, disordering, or packet loss that occurs during transmission. For every byte transmitted the sequence number must be incremented. In the first two steps of the 3-way handshake, both computers exchange an initial sequence number (ISN). This number can be arbitrary, and should in fact be unpredictable, in order to avoid a TCP Sequence Prediction Attack.
TCP primarily uses a cumulative acknowledgment scheme, where the receiver sends an acknowledgment signifying that the receiver has received all data preceding the acknowledged sequence number. Essentially, the first data byte in a segment is assigned a sequence number, which is inserted in the sequence number field, and the receiver sends an acknowledgment specifying the sequence number of the next byte they expect to receive. For example, if computer A sends 4 bytes with a sequence number of 100 (conceptually, the four bytes would have a sequence number of 100, 101, 102, & 103 assigned) then the receiver would send back an acknowledgment of 104 since that is the next byte it expects to receive in the next packet. By sending an acknowledgment of 104, the receiver is signaling that it received bytes 100, 101, 102, & 103 correctly. If, by some chance, the last two bytes were corrupted then an acknowledgment value of 102 would be sent since 100 & 101 were received successfully.
In addition to cumulative acknowledgments, TCP receivers can also send selective acknowledgments to provide further information (see selective acknowledgments).
If the sender infers that data has been lost in the network, it retransmits the data.
The TCP checksum is a quite weak check by modern standards. Data Link Layers with high bit error rates may require additional link error correction/detection capabilities. If TCP were to be redesigned today, it would most probably have a 32-bit cyclic redundancy check specified as an error check instead of the current checksum. The weak checksum is partially compensated for by the common use of a CRC or better integrity check at layer 2, below both TCP and IP, such as is used in PPP or the Ethernet frame. However, this does not mean that the 16-bit TCP checksum is redundant: remarkably, introduction of errors in packets between CRC-protected hops is common, but the end-to-end 16-bit TCP checksum catches most of these simple errors . This is the end-to-end principle at work.
TCP uses an end-to-end flow control protocol to avoid having the sender send data too fast for the TCP receiver to reliably receive and process it. Having a mechanism for flow control is essential in an environment where machines of diverse network speeds communicate. For example, when a fast PC sends data to a slow hand-held PDA, the PDA needs to regulate the influx of data, or protocol software would be overrun quickly. Similarly, flow control is essential if the application that is receiving the data is reading it more slowly than the sending application is sending it.
TCP uses a sliding window flow control protocol. In each TCP segment, the receiver specifies in the receive window field the amount of additional received data (in bytes) that it is willing to buffer for the connection. The sending host can send only up to that amount of data before it must wait for an acknowledgment and window update from the receiving host.
When a receiver advertises a window size of 0, the sender stops sending data and starts the persist timer. The persist timer is used to protect TCP from a deadlock situation that could arise if the window size update from the receiver is lost and the receiver has no more data to send while the sender is waiting for the new window size update. When the persist timer expires the TCP sender sends a small packet so that the receiver sends an acknowledgement with the new window size.
If a receiver is processing incoming data in small increments, it may repeatedly advertise a small receive window. This is referred to as the silly window syndrome, since it is inefficient to send only a few bytes of data in a TCP segment, given the relatively large overhead of the TCP header. TCP senders and receivers typically employ flow control logic to specifically avoid repeatedly sending small segments. The sender-side silly window syndrome avoidance logic is referred to as Nagle's algorithm.
Acknowledgments for data sent, or lack of acknowledgments, are used by senders to infer network conditions between the TCP sender and receiver. Coupled with timers, TCP senders and receivers can alter the behavior of the flow of data. This is more generally referred to as congestion control and/or network congestion avoidance.
In addition, senders employ a retransmission timer that is based on the estimated round-trip time (or RTT) between the sender and receiver, as well as the variance in this round trip time. The behavior of this timer is specified in RFC 2988. There are subtleties in the estimation of RTT. For example, senders must be careful when calculating RTT samples for retransmitted packets; typically they use Karn's Algorithm or TCP timestamps (see RFC 1323). These individual RTT samples are then averaged over time to create a Smoothed Round Trip Time (SRTT) using Jacobson's algorithm. This SRTT value is what is finally used as the round-trip time estimate.
Enhancing TCP to reliably handle loss, minimize errors, manage congestion and go fast in very high-speed environments are ongoing areas of research and standards development. As a result, there are a number of TCP congestion avoidance algorithm variations.
The Maximum segment size (MSS) is the largest amount of data, specified in bytes, that TCP is willing to send in a single segment. For best performance, the MSS should be set small enough to avoid IP fragmentation, which can lead to excessive retransmissions if there is packet loss. To try to accomplish this, typically the MSS is negotiated using the MSS option when the TCP connection is established, in which case it is determined by the maximum transmission unit (MTU) size of the data link layer of the networks to which the sender and receiver are directly attached. Furthermore, TCP senders can use Path MTU discovery to infer the minimum MTU along the network path between the sender and receiver, and use this to dynamically adjust the MSS in order to avoid IP fragmentation within the network.
Relying purely on the cumulative acknowledgment scheme employed by the original TCP protocol can lead to inefficiencies when packets are lost. For example, suppose 10,000 bytes are sent in 10 different TCP packets, and the first packet is lost during transmission. In a pure cumulative acknowledgment protocol, the receiver cannot say that it received bytes 1,000 to 9,999 successfully, but failed to receive the first packet, containing bytes 0 to 999. Thus the sender would then have to resend all 10,000 bytes.
In order to solve this problem TCP employs the selective acknowledgment (SACK) option, defined in RFC 2018, which allows the receiver to acknowledge discontiguous blocks of packets that were received correctly, in addition to the sequence number of the last contiguous byte received successively, as in the basic TCP acknowledgment. The acknowledgement can specify a number of SACK blocks, where each SACK block is conveyed by the starting and ending sequence numbers of a contiguous range that the receiver correctly received. In the example above, the receiver would send SACK with sequence numbers 1,000 and 10,000. The sender will thus retransmit only the first packet, bytes 0 to 999.
The SACK option is not mandatory and it is used only if both parties support it. This is negotiated when connection is established. SACK uses the optional part of the TCP header (see TCP segment structure for details). The use of SACK is widespread - all popular TCP stacks support it. Selective acknowledgment is also used in SCTP.
Since the size field cannot be expanded, a scaling factor is used. The TCP window scale option, as defined in RFC 1323, is an option used to increase the maximum window size from 65,535 bytes to 1 Gigabyte. Scaling up to larger window sizes is a part of what is necessary for TCP Tuning.
The window scale option is used only during the TCP 3-way handshake. The window scale value represents the number of bits to left-shift the 16-bit window size field. The window scale value can be set from 0 (no shift) to 14.
Many routers and packet firewalls rewrite the window scaling factor during a transmission. This causes sending and receiving sides to assume different TCP window sizes. The result is non-stable traffic that is very slow. The problem is visible on some sending and receiving sites which are behind the path of broken routers.
TCP timestamps, defined in RFC 1323, help TCP compute the round-trip time between the sender and receiver. Timestamp options include a 4-byte timestamp value, where the sender inserts its current value of its timestamp clock, and a 4-byte echo reply timestamp value, where the receiver generally inserts the most recent timestamp value that it has received. The sender uses the echo reply timestamp in an acknowledgment to compute the total elapsed time since the acknowledged segment was sent.
TCP timestamps are also used to help in the case where TCP sequence numbers encounter their 232 bound and "wrap around" the sequence number space. This scheme is known as Protect Against Wrapped Sequence numbers, or PAWS (see RFC 1323 for details).
Unfortunately, TCP OOB data was not designed for the modern Internet. The urgent pointer only alters the processing on the remote host and doesn't expedite any processing on the network itself. When it gets to the remote host there are two slightly different interpretations of the protocol which means only single bytes of OOB data are reliable. This is assuming it's reliable at all as it's one of the least commonly used protocol elements and tends to be poorly implemented.
Normally, TCP waits for the buffer to exceed the maximum segment size before sending any data. This creates serious delays when the two sides of the connection are exchanging short messages and need to receive the response before continuing. For example, the login sequence at the beginning of a session begins with the short message "Login," and the session cannot make any progress until these five characters have been transmitted and the response has been received. This process can be seriously delayed by TCP's normal behavior.
However, an application can force delivery of octets to the output stream using a push operation provided by TCP to the application layer. This operation also causes TCP to set the PSH flag or control bit to ensure that data will be delivered immediately to the application layer by the receiving transport layer.
In the most extreme cases, for example when a user expects each keystroke to be echoed by the receiving application, the push operation can be used each time a keystroke occurs. More generally, application programs use this function to force output to be sent after writing a character or line of characters. By forcing the data to be sent immediately, delays and wait time are reduced.
A connection can be "half-open", in which case one side has terminated its end, but the other has not. The side that has terminated can no longer send any data into or receive any data from the connection, but the other side can (but generally if it tries, this should result in no acknowledgment and therefore a timeout, or else result in a positive RST, and either way thereby the destruction of the half-open socket).
It is also possible to terminate the connection by a 3-way handshake, when host A sends a FIN and host B replies with a FIN & ACK (merely combines 2 steps into one) and host A replies with an ACK. This is perhaps the most common method.
It is possible for both hosts to send FINs simultaneously then both just have to ACK. This could possibly be considered a 2-way handshake since the FIN/ACK sequence is done in parallel for both directions.
Some host TCP stacks may implement a "half-duplex" close sequence, as Linux or HP-UX do. If such a host actively closes a connection but still has not read all the incoming data the stack already received from the link, this host will send a RST instead of a FIN (Section 18.104.22.168 in RFC 1122). This allows a TCP application to be sure that the remote application has read all the data the former sent - waiting the FIN from the remote side when it will actively close the connection. Unfortunately, the remote TCP stack cannot distinguish between a Connection Aborting RST and this Data Loss RST - both will make the remote stack to throw away all the data it received, but the application still didn't read.
Some application protocols may violate the OSI model layers, using the TCP open/close handshaking for the application protocol open/close handshaking - these may find the RST problem on active close. As an example:
s = connect(remote);
close(s);For a usual program flow like above, a TCP/IP stack like that described above does not guarantee that all the data will arrive to the other application unless the programmer is sure that the remote side will not send anything.
Impersonating a different IP address was possible prior to RFC 1948, when the initial sequence number was easily guessable. That allowed an attacker to blindly send a sequence of packets that the receiver would believe to come from a different IP address, without the need to deploy ARP or routing attacks: it is enough to ensure that the legitimate host of the impersonated IP address is down, or bring it to that condition using denial of service attacks. This is why the initial sequence number is chosen at random.
TCP uses the notion of port numbers to identify sending and receiving application end-points on a host, or Internet sockets. Each side of a TCP connection has an associated 16-bit unsigned port number (0-65535) reserved by the sending or receiving application. Arriving TCP data packets are identified as belonging to a specific TCP connection by its sockets, that is, the combination of source host address, source port, destination host address, and destination port. This means that a server computer can provide several clients with several services simultaneously, as long as a client takes care of initiating any simultaneous connections to one destination port from different source ports.
Port numbers are categorized into three basic categories: well-known, registered, and dynamic/private. The well-known ports are assigned by the Internet Assigned Numbers Authority (IANA) and are typically used by system-level or root processes. Well-known applications running as servers and passively listening for connections typically use these ports. Some examples include: FTP (21), ssh (22),  (23), SMTP (25) and  (80). Registered ports are typically used by end user applications as ephemeral source ports when contacting servers, but they can also identify named services that have been registered by a third party. Dynamic/private ports can also be used by end user applications, but are less commonly so. Dynamic/private ports do not contain any meaning outside of any particular TCP connection.
TCP is a complex and evolving protocol. However, while significant enhancements have been made and proposed over the years, its most basic operation has not changed significantly since its first specification RFC 675 in 1974, and the v4 specification RFC 793, published in September 1981. RFC 1122, Host Requirements for Internet Hosts, clarified a number of TCP protocol implementation requirements. RFC 2581, TCP Congestion Control, one of the most important TCP-related RFCs in recent years, describes updated algorithms to be used in order to avoid undue congestion. In 2001, RFC 3168 was written to describe explicit congestion notification (ECN), a congestion avoidance signalling mechanism.
The original TCP congestion avoidance algorithm was known as "TCP Tahoe", but many alternative algorithms have since been proposed (including TCP Reno, TCP Vegas, FAST TCP, TCP New Reno, and TCP Hybla).
TCP Interactive (iTCP) is a research effort into TCP extensions that allows applications to subscribe to TCP events and register handler components that can launch applications for various purposes, including application assisted congestion control.
One way to overcome the processing power requirements of TCP is to build hardware implementations of it, widely known as TCP Offload Engines (TOE). The main problem of TOEs is that they are hard to integrate into computing systems, requiring extensive changes in the operating system of the computer or device. The first company to develop such a device was Alacritech.
Also for embedded systems, network booting and servers that serve simple requests from huge numbers of clients (e.g. DNS servers) the complexity of TCP can be a problem. Finally some tricks such as transmitting data between two hosts that are both behind NAT (using STUN or similar systems) are far simpler without a relatively complex protocol like TCP in the way.
Generally where TCP is unsuitable the User Datagram Protocol (UDP) is used. This provides the application multiplexing and checksums that TCP does, but does not handle building streams or retransmission giving the application developer the ability to code those in a way suitable for the situation and/or to replace them with other methods like forward error correction or interpolation.
SCTP is another IP protocol that provides reliable stream oriented services not so dissimilar from TCP. It is newer and considerably more complex than TCP so has not yet seen widespread deployment. However, it is especially designed to be used in situations where reliability and near-real-time considerations are important.
Venturi Transport Protocol (VTP) is a patented proprietary protocol that is designed to replace TCP transparently in order to overcome perceived inefficiencies related to wireless data transport.
TCP also has some issues in high bandwidth utilization environments. The TCP congestion avoidance algorithm works very well for ad-hoc environments where it is not known who will be sending data, but if the environment is predictable, a timing based protocol such as ATM can avoid the overhead of the retransmits that TCP needs.
In other words, after appropriate padding, all 16-bit words are added using one's complement arithmetic. The sum is then bitwise complemented and inserted as the checksum field. A pseudo-header that mimics the IPv4 header, used in the checksum computation, is shown in the table below.
|Bit offset||Bits 0–3||4–7||8–15||16–31|
|96||Source port||Destination port|
The source and destination addresses are those of the IPv4 header. The protocol value is 6 for TCP (cf. List of IP protocol numbers). The TCP length field is the length of the TCP header and data.
An IPv6 pseudo-header for computation of the checksum is shown below.
|Bit offset||Bits 0 - 7||8–15||16–23||24–31|
|320||Source port||Destination port|