Transcript old_Ch4

ICS 156: Networking Lab
Magda El Zarki
Professor, ICS
UC, Irvine
Course Outline




Ch 1: Introduction
Ch 2:Bridges
Ch 3:Routers
Ch 4: Transport Protocols
4. Transport Layer Protocols

Introduction

Transport Layer Protocols:
– Transport Control Protocol (TCP) and
– User Datagram Protocol (UDP)
4.1 Introduction
Host
Host
End-to-End Transport
4
Router
Router
Router
4
3
3
3
3
3
2
2
2
2
2
1
1
1
1
1
Subnet 1
Subnet 2
Subnet 3
Subnet 4
4.1.1 TCP and UDP

There are two kinds of Transport Protocols
used in the Internet:
– UDP: User Datagram Protocol
– TCP: Transport Control Protocol

They both operate over IP.

They both use PORT numbers for
connecting to applications.
4.1.2 Port Numbers (1/3)

The concatenation of IP address and Port Number
provides a unique identifier for transferring
information between a source and a destination.
This is known as a socket, the door between an
application and the underlying transport system.

Each application service is allocated a unique port
number:
– FTP - 21
– Telnet - 23
– Domain Name Server - 53, etc.
4.1.2 Port Numbers (2/3)

There are 1023 port numbers available for
application services.

Not all port numbers are well known, in that
case a request is sent to a catch all port
number. The response packet contains the
correct port number.
4.1.2 Port Numbers (3/3)
Port Number
4
Network IP Address
3
3
3
2
2
1
1
2
2
1
1
Physical Address
Subnet 1
4
Subnet 2
4.1.3 TCP vs UDP

The tuple <Source Port, Destination Port> is used
in the addressing of the Transport Data Unit
(TDU).

Most applications that require a reliable transport
mechanism use TCP, otherwise UDP is used
because it is simpler.

UDP is the preferred choice for real-time services
as it does not incorporate a lot of the overhead that
is associated with TCP.
4.2 UDP (1/3)

Uses only 8 bytes of header:
– 2 bytes for source port
– 2 bytes for destination port
– 2 bytes for length of data unit
– 2 bytes for checksum (covers IP addresses and
IP length field included)
4.2 UDP (2/3)

Provides connectionless service - no sequencing,
no end to end error control (only error detection,
includes data), no flow control.

NFS, some routing protocols (e.g., RIP), RTP
(Real-Time Protocol), TFTP, etc., use UDP as it is
simple and fast if reliability is not an issue.

IP will do fragmentation if the UDP packet is too
large for the path (i.e., > path MTU).
4.2 UDP (3/3)

A UDP packet can be as large as the max.
IP packet minus the headers. Generally the
application will limit the packet size.

UDP leaves most of the work to the upper
(application) layer. This can slow things
down, as ACKs, re-ordering, etc., has to be
done by the application.
4.3 TCP (1/2)
32 bits
Source Port
Dest. Port
Sequence Number
Header Length
Checksum
Acknowledgement Number
Advertised Window
Flags
Urgent Pointer
Options
Data
Flags: 1) SYN - for connection set-up
2) FIN - last segment connection tear down
3) URG - urgent data carried
4) ACK - read ack field
5) RESET, 6) PSH - pass data to application asap
Advertised Window: amount of data sender can send.
Urgent Pointer: Points to where urgent data starts in data field
Checksum: for whole segment
4.3 TCP (2/2)

20 byte header (without options)

Header plus data referred to as a transport
segment

Segment size chosen to accommodate the
smallest IP packet size on path to avoid
fragmentation
4.3.1 Connection Establishment
and Tear Down

TCP goes through a 3 phase set-up:
– Connection set-up (3 way hand-shake): SYN,
SYN ACK, ACK
– Data transfer
– Connection tear down: FIN, ACK, FIN, ACK
(two way or one way)
4.3.2 Sequence Numbers and
ACKs

TCP is connection oriented - uses sequence numbers
(always increasing, numbers not consecutive, reflect the
number of bytes not segments!). For example: if TCP
wants to send 2000 bytes and sends them in two segments
of 1000 bytes each, if first segment has sequence no 102,
then sequence number of second segment will be
102+1000= 1102.

The ACKs on the other hand reflect the sequence number
of the next byte that the receiver is expecting. If the
receiver received segment with sequence number 102 of
length 1000, then ACK will use number: 102+ 1000 =
1102.
4.3.3 Out of Order Arrivals

The RFC does not specify what to do with out of
order packets. TCP will only acknowledge ordered
transmissions, but the question is: should it keep
out of order packets in its buffer or just discard
them? The easiest implementation is to discard
and wait for all packets to arrive in order.

An ACK is sent in response to an out of order
packet, it contains the “expected” sequence
number.
4.3.4 Group and Delayed ACKs

Group ACKs: TCP will use group ACKs, also
referred to as delayed ACKs. in otherwords, it will
not ACK each received segment, it will wait for a
short while (differs on each system, 200msecs for
WANs, 20msec for LANs) and then ACK all
segments (max. 2 segments) that it has received so
far.

It does that so that: 1) it can do group ACKing
which is more efficient, and 2) it hopes to be able
to piggyback an ACK on data going in the reverse
direction. If there is data in the buffer that needs to
be sent, it will not do a delayed ACK. Only uses
delayed ACK if send buffer is empty.
4.3.5 Timers & Retransmissions

Uses error detection and recovery - ACKs
and retransmissions. If an ACK is not
received before a timer expires, the segment
is resent.

The retransmission timer is not a static
value. It is calculated based upon current
network status. The timer must be greater
than the roundtrip delay!
4.3.6 Flow Control (1/2)

Sliding window is used as the flow control
technique (matching of sender and receiver data
rates): Each end system will advertise the largest
window it is willing to receive without ACKs, i.e.,
packets that can be outstanding between source
and destination.

This value is dynamic, it changes based upon how
fast the receiver can read the arriving data. It is
always advertised in the messages sent from the
destination to the source. Note however that the
TCP congestion control mechanism does kick in
too and affects the amount of data that can be sent.
4.3.6 Flow Control (2/2)

If an ACK packet comes back with a
receiver window size of “0”, that means that
the destination wants the source to hold off
with transmission. In this case the sending
host can only send segments with 1 byte of
data. This is to keep the connection alive
between the two ends, i.e., keep ACKs
flowing from destination to source
indicating the window size.
4.3.7 Congestion Control

TCP has imbedded congestion control. When ACKs come
back with a delay (i.e. timers expire) the window size is
reduced by the sender. The sender will only send data =
min{congestion window, receiver (or sliding) window}.

TCP also uses something called slow start: The sender
starts off with only sending 1 (in some cases 2) packet,
then doubles that if the ACK comes back before the
retransmit timer expires. It keeps on doubling until it
reaches a threshold then goes into a linear increase (i.e.,
adds one to the window size). If at anytime an ACK is
delayed, it sets the threshold to half the current congestion
window size and starts again with slow start.
4.3.8 Applications and TCP



All applications that are concerned with
reliable transport use TCP
SMTP (email), ftp, RIP, http, etc. use
TCP.
TCP hides the underlying network from
the application: it segments,it re-orders
packets, does error control and handles
flow matching and congestion control.