Transmission Control Protocol

6CCS3INS Internet Systems

2014-15 Toktam Mahmoodi, Department of Informatics, KCL
Transmission Control Protocol (TCP)
 Operates above IP to offer a number of additional
features, and address issues not tackled by IP
 Multiplexing: Two hosts can have multiple simultaneous
'conversations' without getting confused about which data
belongs to which conversation
 Reliability: Data transmitted is checked and re-transmitted
where necessary
 Flow Control: A receiving host/network can only accept and
process a given amount of data at once, so the sender must
control how fast they send data.
 Congestion Control: The network in between source and
destination receive data from multiple sources and can
potentially get congested.

Key features
 Transmission Control Protocol (TCP)
 Multiplexing
 TCP ports
 Flow Control
 Segmentation
 Reliability
 Acknowledgements
 Congestion Control
 Windowing

Outline
How does a host running multiple networked applications
differentiate between messages?
 TCP conceptually divides the communications a host

can be involved with, into ports.

 A port is identified by a number, used in each message

between hosts so that they know which communication

the message belongs to.
 A few ports are reserved for specific application

protocols, e.g.
 FTP: 20/21 SMTP (e-mail): 25
 HTTP: 80 HTTPS: 443

Multiplexing
 32 bits / 4 bytes

16 bits 16 bits

TCP header – source & destination ports

 In using TCP, hosts must establish a connection
 Data can only be sent during the connection
 Requires connection setup/teardown
 This is to ensure
 Reliability
 Resource reservation to ensure quality of service
 Flow/congestion control

Connection-Oriented
 A socket is the combination of a host's IP address and a
port number
137.73.9.232:80

 Every TCP connections is between two sockets, i.e. two

hosts using particular ports

Sockets
 Server: a host that is ready to accept communication on one or
more ports
 The server signals that it is ready to receive connections on a port
by informing the TCP software on the host: a Passive OPEN request
 For example a web server application running on www.google.com
 Client: a host that communicates with a server
 A client signals that it wishes to connect to a server's socket by
sending a message to the server: an Active OPEN request
 For example, a web browser running on your local PC
 Both hosts can potentially be both client and server in the same
connection

Client and server

 Three-way handshake:
 A handshake is a sequence of messages sent between a client
and server to set up the connection before the message data is
sent
 The handshake involves three steps:
 Client sends a synchronise message to the server
 Server sends a message back acknowledging the synchronise,
and giving permission for communication to take place
 Client sends a message acknowledging the
acknowledgement

Connection set-up
3-way handshake

1) SYN

2) SYN + ACK

3) ACK

Three-Way Handshake
 A handshake is also used to close the connection
 When a client has finished sending data and wishes to close the
connection it sends a finalise message
 The server responds with an acknowledgement of the finalise
 Finally, the client responds with an acknowledgement of the
acknowledgement

Connection tear-down
1) FIN

2) ACK

3) ACK

Connection tear-down
 Transmission Control Protocol (TCP)
 Multiplexing
 TCP ports
 Flow Control
 Segmentation
 Reliability
 Acknowledgements
 Congestion Control
 Windowing

Outline
 A host can only receive and process data at a certain
rate
 At some point its buffers get full and it would have to either
overwrite unprocessed data or ignore incoming data
 To resolve this, TCP allows the server to tell the client
how much data it can handle
 The client will then reduce or increase the rate at which it sends
data to match the server

Flow control
 One part of flow control is to split the message to be
sent into separately transmitted segments
 This is like fragmentation in IP, but is independent and
for different reasons:
 In IP, fragments are used to not exceed the limits of the physical
networks and network access layer protocols
 In TCP, segments are used to not exceed the limits of the
receiving host, flow control, and to aid reliability of
communication

Segmentation
 Every byte in a message sent from a client to a server
has a sequence number
 For a given connection and host, there is an initial
sequence number (ISN) for bytes it sends
 A segment will contain all data within a range of sequence
numbers: ISN+a to ISN+b
 Client communicates its ISN in synchronisation

Sequence numbers
 32 bits / 4 bytes

Sequence
Number in
32 bits the SYN
message, is
the ISN.

TCP header – Sequence Number

 A maximum segment size (MSS) for a connection is
maximum amount of data allowed in one segment
transmitted over that connection

 Specified by server to client during synchronisation

handshake

Maximum segment size

 Transmission Control Protocol (TCP)
 Multiplexing
 TCP ports
 Flow Control
 Segmentation
 Reliability
 Acknowledgements
 Congestion Control
 Windowing

Outline
 The main method by which TCP provides reliable
communication is for the server to send an
acknowledgement back to the client for every segment
it receives
 A client uses acknowledgements to determine whether a
segment has been lost
 If it believes a segment is lost, it will re-transmit it

Reliability
 An acknowledgement states that the server has
received all data in the message before a given
sequence number
 If the server has received segments with sequence numbers (1 -
20) and (21 - 30), it will send an acknowledgement with
sequence number 31.
 If the server has received segments with sequence numbers (1 -
20) and (50 - 60), it will send an acknowledgement with
sequence number 21, as it has all data up to there.

Acknowledging sequences
Acknowledgements (ACK)
Acknowledgements
Acknowledgements
Acknowledgements
Acknowledgements
Acknowledgements
Acknowledgements
 Client will REtransmit the lost segment(s) to the server.

 How segment is identified lost by the client?

1) RTO: The client may decide a packet is lost if it has not

received an acknowledgement for it after a given period
of time.
 Retransmission Time Out (RTO)

Retransmission Policy I
2) DupACK: The client may decide packet X is lost, if it
repeatedly receives acknowledgements suggesting
other segments are being received but X has not
(typically 3 times).
 For example, server receives segments (1-20), (30-40), (41-44)
 it will send an acknowledgement with sequence number 31 for 3
times: (1-20), ACK 21, (30-40), ACK 21, (41-44), ACK 21
 The repetition suggests to the client that one or more segments
starting from SN 21 are lost and needs resending.

Retransmission Policy II
 32 bits / 4 bytes

32 bits

TCP header – Acknowledgement

 Transmission Control Protocol (TCP)
 Multiplexing
 TCP ports
 Flow Control
 Segmentation
 Reliability
 Acknowledgements
 Congestion Control
 Windowing

Outline
 Flow control
 end-to-end mechanism for regulating traffic between source and
destination (client and server).

 Congestion control
 Mechanism used by the network to limit congestion

 The two are not really separable.

 In either case, mechanisms are needed for limiting the amount
of traffic entering the network

Congestion Control
 The amount of data that client can send at each time is
called the congestion window size - cwnd
 The client then wait to receive ACK for the sent data and then
send more data.

 The cwnd size will be adjusted during communication if

more/less data is delivered to the server.
 This can be recognised from the rate with which ACK messages
are delivered to the client.

Windowing
 ?

Windowing
 TCP cwnd is initialised to one segment at the start of the
TCP connection, and increases slowly to examine the
available bandwidth.

 The cwnd increase rate depends whether TPC is in

 Slow start, or
 Congestion Avoidance

TCP Congestion Control - cwnd

 TCP congestion control is in Slow Start state at the beginning of the
connection

 In Slow Start,
 cwnd is initialised to one
 cwnd is increased exponentially to quickly capture the available
bandwidth
 Factor of 2 if every packet is ACK’d

TCP Congestion Control – slow start

 TCP Congestion control enters congestion avoidance
state, after cwnd reaches certain threshold
 The threshold is called slow start threshold and shown as
ssthresh

 In Congestion Avoidance
 cwnd is increased by 1/cwnd (linear increase)

TCP Congestion Control – congestion avoidance

 Round Trip Time (RTT) or Round Trip Delay (RTD) is
 the amount of time it takes for an ACK to be received by the
sender
 starting from the time the original packet was sent
 e.g. ping time

 The rate with which cwnd increases in the congestions

control is RTT.
 it’s the clock of TCP.

Round Trip Time

 On detecting a packet loss, client assumes that network
is congested congestion
 Client drastically reduces the congestion window

 Reducing cwnd reduces amount of data that can be sent

per RTT
 Less traffic in the network
 TCP Throughput decreases

TCP Congestion Control – packet loss

 Long-lived TCP session (“sawtooth” curve)
congestion
window

24 Kbytes

16 Kbytes

8 Kbytes

time

TCP congestion window

 If packet loss occurs, TCP
 reduces the cwnd
 retransmits the sequence starting from the lost packet.

 In case of loss detection via DupAck

 cwnd will be reduced to half
 TCP remains in congestion avoidance

 In case of RTO
 ssthresh will be set to half the current cwnd
 cwnd will be reduced to one
 Slow start is initiated

TCP Retransmission Behaviour

Effect of RTO expiry
 32 bits / 4 bytes

Current
cwnd size

16 bits

TCP header – Window

 Offset – 4 bits: Length of the TCP header in 32-bit words.
 Reserved – 6 bits: reserved for future uses ( all 0’s)
 Checksum – 16 bits: this is a one's complement of a one's
complement sum of 16-bit words (similar to IPv4 checksum)
 Urgent Pointer – 16 bits
Position of where the urgent
data ends (as a SN)

TCP Header - more

Flags:
 RST: If a host receives unexpected TCP data it will respond with a
reset message to stop the connection
 PSH: A client can request that all the data ready to send should be
sent. Flag indicates this event to the server
 URG: urgent data
 ACK: acknowledgement
 SYN: synchronisation
message
 FIN: the connection
termination message

TCP Header - Flags

Options:
 Various TCP options - varies in length
 An important TCP option is the MSS - 16 bits
 This is used for the server to specify the largest segment size
it is willing to accept
 It may only be used in the
SYN message.
 The server can declare
the MSS in the message
acknowledging SYN

TCP Header - Options field

 RFCs
 TCP 793
 TCP and IP 879
 Maximum Segment Size 879
 Other issues relevant to TCP 2001

 W. Richard Stevens, "TCP/IP Illustrated, Volume 1: The

Protocols", Addison-Wesley Professional computing
series, 1994
 The TCP/IP guide: http://www.tcpipguide.com/

Relevant RFCs & Extra Reading