Transport Layer Protocols
Version 2.0
Transport layer performs two main tasks for the application layer by using the network layer. It provides end‐to‐end communication between two applications, and implements some control functions. The original TCP/IP protocol suite (to be introduced later) introduces two different protocols for this purpose: Transmission Control Protocol (TCP) and User Datagram Protocol (UDP). TCP establishes a virtual link between two application programs, and provides error checking and congestion control. On the other hand, UDP protocol keeps the minimum requirement of an end‐to‐end connection and leaves the control mechanisms to the application. TCP is a reliable and connection‐oriented protocol appropriate for applications such as email delivery and file transfer protocol (FTP). The UDP is an unreliable and connectionless protocol, but it is simple and incurs less communication overhead.
The UDP is a good choice for applications such as Voice‐over‐IP (VoIP), where packet loss is acceptable to some extent. In the following section, the TCP and UDP are introduced.
Transmission Control Protocol (TCP)
TCP is the main protocol of the transport layer in packet switched networks. TCP
and IP protocols provide the basic structure of the Internet. These two protocols are
complementary in the sense that IP facilitates packet forwarding irrespective of
reliable packet delivery, whereas TCP is responsible for reliable packet delivery
between applications. In the network layer, the order of packets during
transmission may change, because, different packets may reach a destination on
different routes. Moreover, some packets may be dropped due to the congestion
and/or long delay in a node on the path. TCP protocol compensates for some of
To implement reliable services, TCP deploys a numbering system as well as some control mechanisms. The numbering system determines byte, sequence, and acknowledgment numbers. The control mechanisms include flow, error, and congestion controls. The bytes of the data part of TCP segments are numbered by the protocol. In fact, the first byte of a data segment gets a sequence number explicitly, where as the following bytes get it implicitly. The number is put in the sequence number field of the segment header, and is used for flow and error control.
In addition, the parties involved in a connection use the {byte number + 1} in their segment header as acknowledgment number. It indicates the byte number they expect to receive. For example, if the acknowledgement number of a segment header of one of the communication parties shows 2304, that means the party has received all of the bytes from the beginning up to and including byte number 2303, and now it is expecting to receive byte number 2304. TCP connections are full duplex; each party has its own sequence and acknowledgment numbers for the transmitting stream.
Flow control avoids data overflow at the receiver. In other words, the receiver tells the sender how much to send so that buffer overflow is avoided. Congestion control is implemented by the sender based on the level of congestion in a network. Error control mechanism provides a reliable service for TCP. The control mechanisms of TCP will be explained later on. Prior to that, TCP protocol suite is introduced to show the relevance between TCP and the upper and lower layer protocols. Also, the structure of a TCP segment and the concept of connection oriented TCP are described to help understanding the TCP control mechanisms.
TCP/IP Protocol Suite
TCP and IP protocols were introduced prior to the OSI (Open System
protocol stack based on the TCP/IP, as shown in Fig.1, does not exactly match with the 7‐layer OSI model. The TCP/IP protocol suite contains four layers of protocols (or five layers, depending on which reference you look at!) as defined below.
Layer 4 ‐ Application Layer: This layer performs the functionalities of application, presentation and session layers of the OSI model. Application protocols such as SMTP (simple mail transfer protocol), FTP (file transfer protocol), and HTTP (hypertext transfer protocol) operate at this layer.
Layer 3 ‐ Transport Layer: This layer performs the equivalent functions of the OSI transport layer as well as some OSI session layer functionalities. This layer provides different levels of reliable delivery service for the application layer depending on the application requirements.
Layer 2 ‐ Internet Layer: This layer handles routing tables and packet forwarding.
Layer 1 ‐ Network Access Layer (Data Link + MAC + Physical): Connection to the physical environment and flow control are performed at this layer. The layer specifies the characteristics of the hardware and access methods.
Physical Data Link Network Transport Session Presentation
Application
Network Interface
Internet
Transport
Application
OSI TCP/IP
TCP Segment
In order to understand the TCP protocol, it is necessary to understand the various fields of a TCP segment header. These fields are described in this section. In the next sections, we describe how each field is utilized for each TCP service. The TCP Segment header is 20 bytes long, but it may increase up to 60 bytes if an optional 40 byte is added to the header. The format of a segment is shown in Fig. 2. The meaning and purpose of each field is as follows:
Source Port Address: A 16‐bit address of the port number of the sender (application).
Destination port address: A 16‐bit address of the port number of the receiver (application).
Sequence Number: A 32‐bit address (sequence number) of first byte of data in each segment. In other words, the sequence number indicates the address of the first data byte in each segment for the receiver.
Acknowledgment Number: A 32‐bit byte number that the receiver expects to receive from the sender. An acknowledgment number, 1023, means that all the bytes starting from the initial sequence number and up to and including byte number 1022 have been successfully received.
Header length: A 4‐bit field that represents the header length in multiple of four bytes.
Reserved: A 6‐bit field reserved for the requirements of future protocol
developments.
Control: A 6‐bit field contains some flags for connection establishment, termination, abortion, flow control, and data transfer mode indication. Table 1 represents the usage of each flag.
Source Port D estination Port
Sequence N um ber
A cknow ledgm ent Num ber
W indow size H eader
Length
R eserved
U rgent Pointer C hecksum
Options Padding
Data
H e a d e r
U A P R S F
U : U RG (U rgent) A: ACK P: PSH (Push)
R : RST (Reset) S: SY N (Sync.) F: FIN (Finish)
Fig. 2: TCP Segment Format
Table 1‐ TCP Segment Header Flags Flag value Description
URG 1 Urgent pointer field has a valid value.
ACK 1 Acknowledgment field has a valid value.
PSH
1 The receiving TCP module passes (pushes) the data to the application immediately.
0 The receiving TCP module may delay the data.
RST 1 The connection request is denied, or the connection is aborted.
SYN 1 Synchronize sequence number during connection
FIN 1 Sender has no more data to send, but is
ready to receive.
Window Size: A 16‐bit number specifying the number of bytes the receiver is willing to receive. The value of this field is used in flow control and congestion control.
Checksum: A 16‐bit field used for error correction.
Urgent pointer: A 16‐bit number that is added to the sequence number and shows the location of the first byte of urgent data in the data section of the segment. Urgent data is delivered to the recipient application program to be processed before other data. The value of this field is used when the URG flag is on.
Options: The length of this field may reach up to 40 bytes.
Connection oriented TCP
TCP is a connection oriented protocol. It establishes a virtual connection between the running application programs (process) at the two communication end points.
TCP uses port number for this purpose. Each port number is relevant to an application as shown in Table 2. The stream of data is sent from the process to the transport layer. TCP establishes a connection between the transport layers of the transmitter and receiver stations. It then divides the stream of data into units called segments. Segments are numbered and transmitted one by one. The TCP protocol at the receiver side checks the arriving segments for error, loss, and duplication. It orders the segments, makes a stream, and transfers the stream to the receiver side process. TCP provides stream delivery service. Both the sender and the receiver processes deal with the stream of bytes and are not aware of stream segmentation in the lower layers. After all segments are transmitted, TCP at the transmitter side closes the connection.
Table 2‐ Port Numbers Used by TCP Port Protocol Description
20 FTP, Data File Transfer Protocol (data connection) 21 FTP, Control File Transfer Protocol (control
connection) 23 TELNET Terminal Network
25 SMTP Simple Mail Transfer Protocol
53 DNS Domain Name Server
79 Finger Finger
80 HTTP Hypertext Transfer Protocol 111 RPC Remote Procedure Call
Connection oriented TCP consists of connection establishment, data transfer, and connection termination phases. The parties involved in a connection establishment are termed as client, applicant for the connection, and server, the destination.
Connection establishment is a three‐way handshaking protocol. Fig. 3 shows a connection establishment. First, the client sends a segment with SYN=1 to synchronize the sequence numbers. The segment contains the initial sequence number. Second, the server sends the second segment with SYN and ACK flags on.
That means the segment contains the initial sequence number of the server and the first segment transmitted by the client has been successfully received. The server’s segment contains the value of the window size kept by the server. In the next sections we will describe the application of the window size. Third, the client sends the segment with ACK flag on which acknowledges the reception of the server segment reception. The window size of the client is included too. Note that the sequence number of the client’s acknowledgment segment is the same as client’s first segment (synchronizing segment), because no data is carried in the first segment and sequence number does not change.
Client Server
Seq# = 8000 , SYN=1
Seq#=150 00, Ack = 8001, SYN=AC K=1, RW ND = 500 0
Seq#=8000, A ck = 15001, ACK=1, RW ND = 10000
Tim e Ti m e
Data transfer phase is bidirectional. It includes data and acknowledgement transmission. The acknowledgment is piggybacked with data or transmitted individually. Fig. 4 shows an example of data transfer. The client and the server have 2000 bytes of data (each one) to transmit. The client transmits the data in two segments, and the server transmits its data all at once. Acknowledgment is piggy backed with the three first segments. However, there is no more data to be transmitted at the last segment, and it carries only the acknowledgement. In the client’s data segments, the PSH flag is on. It means the server TCP should deliver data to server application as soon as they are received (appropriate for interactive
applications).
Fig. 3 – Three‐way connection establishment
Client or server can terminate a connection, no matter which one has established it.
Also, one party can terminate its connection, while it still receives data from the
other party. The termination phase is implemented by terminating party which
transmits a segment contains FIN flag on, with the last available chunk of data. The
recipient party acknowledges the connection termination by activating the ACK flag in its next segment.
Client Server
Seq# = 8001 , Ack Data bytes: 8001-9 =15001, ACK=PS 000 H=1
Seq#=150 01, Ack = 10001, A CK=1 Data byte s: 15001- 17000
Seq#=10000, A ACK=1, RW ND = 10000 ck = 17001,
Tim e Tim e
Seq# = 9001 , Ack Data bytes: 9001-1 =15001, ACK=PSH=1 0000
Fig. 4 – Data transfer
Flow Control
Flow control determines how much data can be transmitted before receiving an
acknowledgment. There are two extremes for this purpose. One asks an
acknowledgment after each one byte of data transmission, which causes delay and
transmission overhead; the other asks for only one acknowledgment after
transmission all of the data, which causes late feedback to the transmitter to
compensate for transmission problems. For instance, if the path is congested,
transmitter will not be aware of that to slow down the transmission rate. TCP flow
control chooses a dynamic approach between these two extremes.
TCP flow control is window‐based. Window size specifies the amount of data that can be transmitted before receiving the acknowledgment. Window slides over the segments as shown in Fig 5. The size of the window varies according to the congestion status of a network, congestion window (CWND), and the receiver buffer size, receiver window (RWND). The congestion window size is determined by the network to avoid congestion. The receiver window indicates the number of bytes that the receiver can accept before its buffer overflows. The value of the receiver window is sent to the transmitter in the acknowledgment message. The size of the sliding window, shown in Fig. 5, is equal to the min (CWND, RWND). Accordingly, the sliding window resizes by opening from the right and closing from the left.
Opening lets more bytes come to the window and closing lets the acknowledged bytes to go out of the window.
B yte stream
Sent
&
AC Ked bytes
Sent
& not A C K ed bytes
W aiting to be sent
N ot in the windo w S liding windo w
Fig. 5 – TCP sliding window
Example: In the example, shown in Fig. 6, the CWND is 20 and RWND is 9 bytes.
Accordingly, TCP window size equals to 9. The bytes up to 2002 have been sent.
Bytes number 2000, 2001, and 2002 has not been acknowledged by the receiver,
but the sender can still transmit bytes 2003 up to 2008.
C an ’t b e s e n t un til w ind o w op e ns S en t an d A C K ed
1 9 9 9
2 0 0 9 2
0 0 2 2 0 0 0
2 0 0 3 2 0 0 1
2 0 0 8 2 0 0 7 2 0 0 6 2 0 0 5 2 0 0 4 N e xt b yte to b e se nt
W in d o w s iz e = m in (20 ,9 )
Fig. 6 ‐ Sliding Window
If the sender receives an acknowledgment value for byte 2003 along with RWND = 9, the window will be updated. It is closed from the left and opened from the right.
The new TCP window will contain bytes 2003 up to 2011.
Question: What happens if the sender or transmitter is very slow in a sliding window mechanism?
Answer: The Silly Window Syndrome occurs! The slow transaction of data causes the window size to be reduced down to one byte. Comparing the overhead of the segment, 40 bytes, with the body, one byte, shows flow control with sliding window mechanism is not efficient at all.
Remedy: If the slowness of the sender’s application causes Silly Window Syndrome, Nagle’s Algorithm is the remedy. The prescription is as follows:
• Sender sends the first segment even if it is a small one
• Next, the sender waits until an ACK is received, or a maximum‐size segment
• Step 2 is repeated for the rest of the transmission.
If the slowness of the receiver’s application causes Silly Window Syndrome, two prescriptions are advised:
Clark’s solution: Send an ACK as soon as the data arrives, and close the window until another segment can be received or buffer is ½ empty.
Delayed ACK: Delay sending the ACK, at most 500 ms; this causes the sender stop transmitting, but it does not change its window size.
Error Control
The reliability of TCP protocol is achieved by error control mechanisms that detect and correct errors. Error detection includes identifying corrupted, lost, out‐of‐
order, and duplicated segments. TCP deploys three tools for error detection and correction: checksum, acknowledgement, and time‐out.
Checksum
Checksum is a simple method for detecting corrupted segments. Each segment contains a 16‐bit field in its header for checksum. Destination TCP discards segments that have checksum error.
Acknowledgement
Every data or control segment with sequence number is acknowledged to confirm
the segment reception. There are two kinds of acknowledgement: positive
acknowledgement (shortly ACK) and selective acknowledgement (SACK). Positive
acknowledgement symbolizes the fact that the receiver advertises the expecting
complementary mechanism for ACK that reports erroneous segments, such as duplicated and out‐of‐order segments, to the sender.
Timeout Retransmission
This is a very important task of error control mechanism. Corrupted, lost, and delayed segments are retransmitted. The criteria for the retransmission are retransmission timer expiration or three duplicate ACK receptions.
The TCP sender starts a retransmission time‐out (RTO) timer when it sends a segment. The segment is retransmitted upon RTO timer expiration, no matter what the reason is for not receiving ACK. For example if the segment is received successfully, but the ACK is lost or delayed, the TCP sender assumes segment has been lost. For this reason, the RTO timer is set for a longer time in networks with long round trip time (RTT). RTT is the time for a segment to catch the receiver plus the time for an ACK to reach to the segment transmitter.
Network that needs fast retransmission, three duplicate ACK is the rule for retransmission. If the sender receives three ACKs for a sequence number, it sends the segment with that sequence number immediately, irrespective of the RTO timer value. The advantage of three duplicate ACK rule is the receiver does not need to buffer many segments until it finally receives an out‐of‐order segment.
Congestion Control
In a network, data are queued in the buffer of the interfaces and delivered to the next stage in an appropriate time. If there is a mismatch between processing time or capacity of the two interconnected stages of the network, congestion occurs.
Congestion affects two performance metrics in a network: throughput and delay.
Delay can be considered as the sum of the propagation, processing, and queuing delay. Indeed, congestion affects queuing delay much more than the others. When the network load increases, congestion deteriorates the delay performance. This situation is depicted in Fig. 7. Again, increased delay worsens the congestion state.
Because the sender attempts to retransmit a packet upon late ACK, which increases the network load further. Throughput and goodput are two performance measures that show the behaviour of the network versus load. Throughput is defined as the number of packets passing through the network in a unit of time. The measure of the received packets to the transmitted packets named goodput. As shown in Fig. 7, when the network load increases up to the network capacity, the throughput and goodput increase. However, when the traffic load is more than the network capacity, overflow occurs in the buffers and packets are dropped.
Network Load
De la y
congestion No
congestion
Ne tw ork c ap acit y
Total Input rate
T o ta l O u tp u t r a te
congestion No
congestion
N et w ork capa cit y
Fig. 7‐ Network performance upon congestion
The principles of congestion control mechanisms are to prevent or remove congestion. Congestion prevention mechanisms control congestion by adopting a good retransmission, acknowledgement, and discard policy. Once congestion occurred, different signalling mechanisms are used by the routers to inform the sender or receiver of the congestion to compensate for that.
Congestion can be avoided by appropriately adjusting the retransmission timers
acknowledgement affects the sender’s transmission speed. So, by controlling this time, congestion can be controlled. For instance, if the receiver sends the acknowledgements later, but before the expiration of the sender’s timer, the arrival traffic in a network slows down. Discard policy can prevent congestion by discarding the packets that have less effect on the transmission or application quality, e.g. discarding less important voice or video packets instead of an FTP session data packets. The above preventing mechanisms are open loop control mechanisms and can be implemented at sender or receiver side.
To alleviate an occurred congestion, closed loop control mechanisms are used to slow down the arrival of traffic to a network. When a router experience congestion, it may ask the previous routers to slow down their packet transmissions. Inform directly senders or receivers, instead of routers on a path, to adjust the injected amount of data to the network could be another mechanism. In other words, congestion is controlled by TCP senders and receivers, not routers on a path. The main tool of closed loop congestion control mechanisms is window size. The window size is determined by the minimum value of available capacity of buffers in TCP receiver (RWND) or congestion level in the network (CWND). RWND is advertised from the receiver side and used for flow control. CWND size is determined on sender side and used for congestion control.
TCP Congestion Policy
TCP congestion control mechanism has three instruments: slow start, congestion
avoidance, and congestion detection. A sender starts transmission with a slow rate
and increases the rate as long as its rate is below a threshold. When the threshold is
reached, the rate is decreased to avoid congestion. If congestion happens, the sender
slows down the rate to the slow start rate or another rate of congestion avoidance
phase depending on the chosen policy.
In slow start phase, a sender chooses the minimum value of CWND which has been agreed upon connection establishment. Suppose RWND is much higher than the CWND. After transmitting the first segment and receiving the corresponding ACK, sender doubles CWND. That means two segments can be transmitted before receiving an acknowledgement. If the sender receives ACK for the two transmitted segments, it doubles CWND again, as shown in Fig. 8. The process of increasing CWND exponentially stops and congestion avoidance phase starts when CWND size reaches a threshold value. CWND size increases in congestion avoidance phase linearly, not exponentially, as long as no congestion is detected. That means the sender increases CWND by one after receiving ACK, as shown in Fig. 9. The slow increment of CWND prevents congestion to some extent, but it can not 100% avoid that.
Sender Receiver
Time Time
ACK 2
ACK 4
ACK 8 Segment 1
Segment 2
Segment 3
Segment 4 Segment 5 Segment 6 Segment 7 cwnd=1
cwnd = 1x2 = 21 = 2
cwnd = 2x2 = 22 = 4
cwnd = 4x2 = 23 = 8
Fig. 8‐ slow start phase
A TCP sender assumes congestion happened upon RTO expiration or receiving three
ACKs for a segment. A new slow start or congestion avoidance phase follows the
congestion detection phase. If detection is based on time‐out, the threshold value
will be set to the half of the current window size and CWND will be set by the
minimum CWND size of the slow start phase. Otherwise, if detection is by three
ACKs, the threshold value will be set to the half of the current window size, CWND will be set by the threshold value, and a congestion avoidance phase starts. The different reaction to congestion detection is because of the fact that there is a stronger possibility of congestion when a time‐out occurs than three ACKs occur. In the former, there is no idea about the transmitted segments, but the later means some segments have reached the receiver successfully.
Sender Receiver
Time Time
ACK 2
ACK 4
ACK 7 Segment 1
Segment 2 Segment 3
Segment 4 Segment 5 Segment 6 cwnd=1
cwnd = 1+1 = 2
cwnd = 2+1 = 3
cwnd = 3+1 =4
