Computer Networks

Chapter 06
(version June 8, 2007)

Maarten van Steen

Vrije Universiteit Amsterdam, Faculty of Sciences
Dept. Computer Science
Room R4.20. Tel: (020) 598 7784
E-mail: [email protected], URL: www.cs.vu.nl/∼steen/

01 Introduction
02 Physical Layer
03 Data Link Layer
04 MAC Sublayer
05 Network Layer
06 Transport Layer
07 Application Layer
08 Network Security

00 – 1 /

Transport Layer (1/2)

Essence: The transport layer is responsible for com-

pleting the services of the underlying network to the
extent that application development can take place:

• provide reliable connection-oriented services

• provide unreliable connectionless services

• provide parameters for specifying quality of ser-

vices

Important: we’re talking about efficient and cost-effective

services, in particular reliable connections.

Note: depending on the services offered by the net-

work layer, the added functionality in the transport layer
can vary considerably.

06 – 1 Transport Layer/6.1 Transport Service

 Transport Layer (2/2)
Host 1 Host 2

Application Application
(or session) Application/transport (or session)
layer Transport interface layer
address

TPDU
Transport Transport
entity Transport entity
protocol

Network
address Transport/network
Network layer interface Network layer

Note: The issue here is that the network layer is in the

hands of carriers: organizations that offer a (generally
wide-area) computer network to their clients. Clients
have no say in what the carrier actually offers.

Consequence: If we want to develop applications

that are independent of the particular services offered
by a carrier, we’ll have to devise a standard commu-
nication interface and implement that interface at the
client’s sites. The transport layer contains such imple-
mentations.
06 – 2 Transport Layer/6.1 Transport Service

Transport Layer Interface

Example: Consider the Berkeley socket interface,

which has been adopted by all UNIX systems, as well
as Windows:
SOCKET Create a new communication endpoint
BIND Attach a local address to a socket
LISTEN Announce willingness to accept N con-
nections
ACCEPT Block until someone remote wants to
establish a connection
CONNECT Attempt to establish a connection
SEND Send data over a connection
RECEIVE Receive data over a connection
CLOSE Release the connection

06 – 3 Transport Layer/6.1 Transport Service

 Socket Communication

• The client and server each bind a transport-level

address and a name to the locally created socket.

• The server must listen to its socket, thereby telling

the kernel that it will subsequently wait for con-
nections from clients.

• After that, the server can accept or select con-

nections from clients.

• The client connects to the socket. It needs to

provide the transport-level address by which it can
locate the server.

After a connection has been accepted (or selected),

the client and server communicate through read/write
operations on their respective sockets.

Communication ends when a connection is closed.

06 – 4 Transport Layer/6.1 Transport Service

Connection-Oriented Socket
Communication
SERVER CLIENT

socket() socket()

bind()

listen()

accept() connect()

write()
read()

write()
read()

close() close()

Question: What about connectionless communica-

tion?

06 – 5 Transport Layer/6.1 Transport Service

 Sockets – Server Side
TransportAddress serverAddress /* Publicly known address */
...
void server() {
Socket clientSocket; /* Private socket */
...
serverSocket = new Socket();
serverSocket.bind(serverAddress);
serverSocket.listen(maxConnections);
while( TRUE ) {
serverSocket.accept(&clientSocket);
clientSocket.read(&request); /* receive */
clientSocket.write(answer); /* send */
clientSocket.close();
}
}

06 – 6 Transport Layer/6.1 Transport Service

Sockets – Client Side

TransportAddress serverAddress /* Publicly known address */
...
void client() {
TransportAddress clientAddress; /* Private address */
Socket clientSocket; /* Private socket */
...
clientAddress = new TransportAddress();
clientSocket = new Socket();
connected = FALSE;
clientSocket.bind(clientAddress);
while( !clientSocket.connect(serverAddress) )
sleep(1);
clientSocket.write(&request); /* send */
clientSocket.read(answer); /* read */
clientSocket.close();
}

Question: What am I doing in the loop?

06 – 7 Transport Layer/6.1 Transport Service

 Some Observations
Note 1: Messages sent by clients are encapsulated
as transport protocol data units (TPDUs) to the net-
work layer:
Frame Packet TPDU
header header header

TPDU payload

Packet payload

Frame payload

Note 2: A real hard part is establishing and releasing

connections. The model can be either symmetric or
asymmetric:

Symmetric: one side sends a disconnect request, and

waits for the other to acknowledge that the con-
nection is closed. Yes, there are some problems
with this model. In fact, it turns out it is impossible
to implement.
Asymmetric: one side just closes the connection, and
that’s it. Yes, it’s simple, but you may lose some
data this way. Not really acceptable.

06 – 8 Transport Layer/6.1 Transport Service

Sockets – State Diagram

Connection request Connect primitive
TPDU received executed
IDLE

PASSIVE ACTIVE
ESTABLISHMENT ESTABLISHMENT
PENDING PENDING

Connect primitive Connection accepted

executed TPDU received
ESTABLISHED

Disconnection Disconnect
request TPDU primitive
PASSIVE received ACTIVE
executed
DISCONNECT DISCONNECT
PENDING PENDING

IDLE
Disconnect Disconnection request
primitive executed TPDU received

Note: Dashed lines are server state transitions; solid

lines client state transitions.

06 – 9 Transport Layer/6.1 Transport Service

 Transport Protocol

Observation: transport protocols strongly resemble

those in the data link layer: e.g. lots of error and flow
control. Big differences when it comes to solutions!

Router Router Subnet

Physical
communication channel Host

(a) (b)

• explicit addressing

• establishing, maintaining, and releasing connec-

tions

• the many connections require different solutions

• handle effects of subnet storage capabilities

06 – 10 Transport Layer/6.2 Transport Protocol Elements

Addressing
Note: Each layer has its own way of dealing with ad-
dresses. In IP, a transport service access point is
an IP address with a port number.

Host 1 Host 2
Server 1 Server 2
Application TSAP 1208 Application
process layer

Transport
Transport layer TSAP 1522 TSAP1836
connection

NSAP Network
layer NSAP

Data link
layer

Physical
layer

Question: How do we get to know where the other

party is?
06 – 11 Transport Layer/6.2 Transport Protocol Elements
 Service Locations
Fixed Addresses

General solution: have a single process, located at

a well-known address, handle a large number of ser-
vices (inetd in the UNIX world):

Host 1 Host 2 Host 1 Host 2

Time-
of-day
Layer server

User Process User Process

Server Server

4
TSAP

(a) (b)

06 – 12 Transport Layer/6.2 Transport Protocol Elements

Service Locations
Unknown Addresses

Problem: Sometimes you just can’t have a process

handle all services, e.g. because the service requires
special hardware (file server) ⇒ find address of the
server

Solution: you’ll have to use a name server.

Question: Great, so how do we find the name server?

Next problem: A name server returns a TSAP, not

an NSAP. So how do we get to know the network ad-
dress?

Question: At what level do you think name servers fit

in?

06 – 13 Transport Layer/6.2 Transport Protocol Elements

 Connection Establishment
Basic idea: To establish a connection, you send off
a connection request to the other end. The other end
then accepts the connection, and returns an acknowl-
edgment.

Big problem: Suppose you don’t get an answer, so

you do another request.

• Your first request didn’t make it: no harm done.

• The ack didn’t make it back: you’re establishing a

second connection, but this can probably be de-
tected.

• Your first request didn’t make it yet: now you’re

really making a second connection and no one
knows you didn’t do this on purpose.

Main cause: The network has storage capabilities,

and unpredictable delays. This means that things can
pop up out of the blue.
06 – 14 Transport Layer/6.2 Transport Protocol Elements

Attacking Duplicates (1/2)

Solution: Restrict the lifetime of TPDUs – if the max-
imum lifetime is known in advance, we can be sure
that a previous packet is discarded and that it won’t
interfere with successive ones.

Basic idea: Assign sequence numbers to TPDUs,

and let the sequence number space be so large that
no two outstanding TPDUs can have the same num-
ber.

Problem: When a host crashes, it has to start num-

bering TPDUs again. So, where does it start?

• You can’t just wait the maximum packet lifetime T

and start counting from the start again: in wide-
area systems, T may be too large to do that.

• The point is that you must avoid that an initial se-

quence number corresponds to a TPDU still float-
ing around. So, just find the right initial number.

06 – 15 Transport Layer/6.2 Transport Protocol Elements

 Attacking Duplicates (2/2)
Solution: Assign sequence numbers in accordance
to clock ticks, and assume that the clock continues
ticking during a crash. This leads to a forbidden re-
gion:

2k–1
Forbidden T T
message
Sequence numbers

120
Sequence numbers
n
on de
gi id
re orb
F

80
70
Restart after
60 crash with 70

Actual sequence
numbers used
0
0 30 60 90 120 150 180
Time Time
(a) (b)

Every time you want to assign a next sequence num-

ber, check whether that number is in the forbidden re-
gion.

Watch it: when sequence numbers are assigned at

a lower pace than the clock ticks, we may enter the
region “from the top.” Likewise, assigning them too
fast makes you enter the region “from the bottom.”
06 – 16 Transport Layer/6.2 Transport Protocol Elements

Error-Free Connection
Establishment (1/2)

Problem: Great, we have a way of avoiding dupli-

cates, but how do we get a connection in the first
place?

Note: One way or the other we have to get the sender

and receiver to agree on initial sequence numbers.
We need to avoid that an old (unnumbered) connec-
tion request pops up.

06 – 17 Transport Layer/6.2 Transport Protocol Elements

 Error-free Connection
Establishment (2/2)
Solution: Three-way handshake.

Host 1 Host 2 Host 1 Host 2

Old duplicate
CR (s
eq = CR (s
x) eq = x)

= x) = x)
ACK ACK
Time

= y, eq = y,
eq
ACK (s AC K (s

DAT
A (s REJEC
eq = T (AC
x, AC K = y)
K=
y)

(a) (b)

Host 1 Host 2
CR (s
eq =
x)
Old duplicate
x )
CK=
,A
=y
seq
K(
AC
DAT
A (s
ACK eq = x,
= z)
Old duplicate

REJ
ECT
(ACK
= y)

(c)

06 – 18 Transport Layer/6.2 Transport Protocol Elements

Error-free Connection Release

Asymmetric release: one party just closes down the

connection. May result in loss of data:

Host 1 Host 2

CR

ACK

DATA
Time

DATA

DR
No data are
delivered after
a disconnect
request

06 – 19 Transport Layer/6.2 Transport Protocol Elements

 Symmetric Connection Release
(1/2)
Big problem: Can we devise a solution to release a
connection such that the two parties will always agree.
The answer is simple: NO.

• Normal case: Host 1 sends disconnect request

(DR). Host 2 responds with a DR. Host 1 acknowl-
edges, and ACK arrives at host 2.
• ACK is lost: What should host 2 do? It doesn’t
know for sure that its DR came through.
• Host 2’s DR is lost: What should host 1 do? Of
course, send another DR, but this brings us back
to the normal case. This still means that the ACK
sent by host 1 may still get lost.

Pragmatic solution: Use timeout mechanisms. This

will catch most cases, but it is never a fool-proof solu-
tion: the initial DR and all retransmissions may still be
lost, resulting in a half-open connection.
06 – 20 Transport Layer/6.2 Transport Protocol Elements

Symmetric Connection Release

(2/2)
Host 1 Host 2 Host 1 Host 2
Send DR DR Send DR DR
+ start timer + start timer

Send DR Send DR
+ start timer
DR DR + start timer
Release Release
connection connection

Send ACK ACK

ACK
Release Send ACK
connection Lost (Timeout)
release
connection

(a) (b)

Host 1 Host 2 Host 1 Host 2

Send DR DR Send DR DR
+ start timer Send DR & + start timer Send DR &
DR start timer start timer
Lost Lost
( Timeout)
DR
send DR ( Timeout)
Send DR &
+ start timer send DR
start timer Lost
DR + start timer
Release
connection

Send ACK ACK

Release (N Timeouts) (Timeout)
connection release release
connection connection

(c) (d)

06 – 21 Transport Layer/6.2 Transport Protocol Elements

 Flow Control and Buffering
Main problem: Hosts may have so many connections
that it becomes infeasible to allocate a fixed number
of buffers per connection to implement a proper sliding
window protocol ⇒ we need a dynamic buffer alloca-
tion scheme.

• With an unreliable network, i.e. unreliable data-

gram service provided by the network layer, the
sender will have to buffer TPDUs until they are
acknowledged.
• The receiver may decide to drop incoming TPDUs
if it has no buffer space available.
• With a reliable network, the sender will still have
to buffer a TPDU until it is acknowledged: the net-
work layer cannot help here! (WHY NOT?)

In general: the sender and receiver need to negotiate

the number of TPDUs that can be transmitted in se-
quence, only because buffer space no longer comes
for free.
06 – 22 Transport Layer/6.2 Transport Protocol Elements

Buffer Reservation
Basic idea: The sender requests a number of buffers
at the receiver’s side when opening a connection. The
receiver responds with a credit grant. After that, the
receiver grants more credit when bufferspace becomes
available:
A Message B Comments

1 < request 8 buffers> A wants 8 buffers

2 <ack = 15, buf = 4> B grants messages 0-3 only
3 <seq = 0, data = m0> A has 3 buffers left now
4 <seq = 1, data = m1> A has 2 buffers left now
5 <seq = 2, data = m2> Message lost but A thinks it has 1 left
6 <ack = 1, buf = 3> B acknowledges 0 and 1, permits 2-4
7 <seq = 3, data = m3> A has 1 buffer left
8 <seq = 4, data = m4> A has 0 buffers left, and must stop
9 <seq = 2, data = m2> A times out and retransmits
10 <ack = 4, buf = 0> Everything acknowledged, but A still blocked
11 <ack = 4, buf = 1> A may now send 5
12 <ack = 4, buf = 2> B found a new buffer somewhere
13 <seq = 5, data = m5> A has 1 buffer left
14 <seq = 6, data = m6> A is now blocked again
15 <ack = 6, buf = 0> A is still blocked
16 <ack = 6, buf = 4> Potential deadlock

Question: what can we do about the potential dead-

lock?
06 – 23 Transport Layer/6.2 Transport Protocol Elements
 Flow Control – The Network

Problem: Now that we’ve adjusted the transmission

rate between the sender and receiver, let’s consider
the network capacity as well: it may not be enough for
what the sender and receiver want to do.

Issue: If the network can handle c TPDUs per sec-

ond, and takes a total of r seconds to transmit, prop-
agate, queue and process the TPDU, and to send an
ACK, the sender need only maintain c · r buffers. More
buffers is overkill of the network.

Solution: Let the sender estimate c and r (HOW?)

and adjust its own number of buffers.

06 – 24 Transport Layer/6.2 Transport Protocol Elements

Multiplexing
Basic idea 1: Assuming that the network offers only a
limited number of virtual circuits, or that a user doesn’t
want to pay so much, then use a single circuit for sev-
eral connections ⇒ upward multiplexing.

Basic idea 2: If a user requires a lot of bandwidth

that cannot be supported by a single network virtual
circuit, use several circuits for a single connection ⇒
downward multiplexing.

Transport address
Layer
Network
4 address

2 Router lines

To router
(a) (b)

06 – 25 Transport Layer/6.2 Transport Protocol Elements

 Crash Recovery (1/2)

Problem: A host responds to the receipt of a TPDU

by performing an operation and returning an acknowl-
edgment. How should the sending host respond when
the receiving host crashes before, during, or after its
response?

06 – 26 Transport Layer/6.2 Transport Protocol Elements

Crash Recovery (2/2)

Situation: Assume the sender is informed that the

receiver has just recovered from a crash. Should the
sender retransmit the TPDU it just sent, or not? Dis-
tinguish between:

• S0: sender had no outstanding (unacknowledged)

TPDUs
• S1: sender had one outstanding TPDU

Strategy used by receiving host

First ACK, then write First write, then ACK

Strategy used by
sending host AC(W) AWC C(AW) C(WA) W AC WC(A)

Always retransmit OK DUP OK OK DUP DUP

Never retransmit LOST OK LOST LOST OK OK

Retransmit in S0 OK DUP LOST LOST DUP OK

Retransmit in S1 LOST OK OK OK OK DUP

OK = Protocol functions correctly

DUP = Protocol generates a duplicate message
LOST = Protocol loses a message

06 – 27 Transport Layer/6.2 Transport Protocol Elements

 Example Protocol
Service Primitives
LISTEN Block connection request comes in
CONNECT Attempt to establish a connection
SEND Send data over a connection
RECEIVE Receive data over a connection
DISCONNECT Release the connection

Network Layer Packets

CALL REQUEST Sent to establish a connection
CALL ACCEPTED Response to CALL REQUEST
CLEAR REQUEST Sent to release a connection
CLEAR CONFIRM Response to CLEAR CONNECTION
DATA Used to transport data
CREDIT For managing the window

State of a Connection
IDLE Not yet established
WAITING CONNECT called; CALL REQ. sent
QUEUED CALL REQ. arrived; LISTEN not called
ESTABLISHED Connection established
SENDING Waiting for permission to send TPDU
RECEIVING RECEIVE has just been called
DISCONNECTING DISCONNECT ahs just been called

06 – 28 Transport Layer/6.3 Example Protocol

Example Protocol – FSM (1/2)

State
Dis-
Idle Waiting Queued Established Sending Receiving connecting
P1: ~/Idle
LISTEN P2: A1/Estab ∼/Estab
P2: A2/Idle
P1: ~/Idle
CONNECT P1: A3/Wait
Primitives

P4: A5/Idle
DISCONNECT P4: A6/Disc

P5: A7/Estab
SEND P5: A8/Send

RECEIVE A9/Receiving

P3: A1/Estab
Call_req P3: A4/Queu'd

Call_acc ∼/Estab
Incoming packets

Clear_req ∼/Idle A10/Estab A10/Estab A10/Estab ∼/Idle

Clear_conf ∼/Idle

DataPkt A12/Estab

Credit A11/Estab A7/Estab

Clock

Timeout ∼/Idle

Predicates Actions
P1: Connection table full A1: Send Call_acc A7: Send message
P2: Call_req pending A2: Wait for Call_req A8: Wait for credit
P3: LISTEN pending A3: Send Call_req A9: Send credit
P4: Clear_req pending A4: Start timer A10: Set Clr_req_received flag
P5: Credit available A5: Send Clear_conf A11: Record credit
A6: Send Clear_req A12: Accept message

06 – 29 Transport Layer/6.3 Example Protocol

 Example Protocol – FSM (2/2)

CONNECT TIMEOUT

IDLE

CLEAR REQ CALL REQ

DISCONNECT
CALL REQ
LISTEN,

WAITING QUEUED

CALL ACC LISTEN

ESTAB-
CREDIT, LISHED RECEIVE
CLEAR REQ
DISCONNECT

SEND DATA,
SENDING RECEIVING
CLEAR REQ

DISCON-
NECTING

CLEAR REQ, CLEAR CONF

06 – 30 Transport Layer/6.3 Example Protocol

UDP

Essence: The User Datagram Protocol is essentially

just a transport-level version of IP.

32 Bits

Source port Destination port

UDP length UDP checksum

Observation: UDP is simple: no flow control, no error

control, no retransmissions

Question: So why not use IP instead?

06 – 31 Transport Layer/6.4 UDP

 RPC
Observation: UDP is widely used for simple client-
server communication in which a procedure is made
available to remote clients (Remote Procedure Call).
The call (including its parameters) is shipped to the
server:
Client CPU Server CPU

Client Server
1 5
stub stub
Client
Server

2 4
Operating system Operating system

Network

1. Client calls the procedure at a local stub

2. Client stub marshalls request: it puts everything
into a (UDP) message
3. The message is transferred over the network
4. The server stub unmarshalls the message...
5. ... and calls the local implementation of the pro-
cedure.
Question: What’s the difficulty with RPCs?
06 – 32 Transport Layer/6.4 UDP

RTP
Problem: Can we support multimedia streaming over
the Internet? The Real-Time Transport Protocol pro-
vides some best-effort support.
Ethernet IP UDP RTP
Multimedia application header header header header
User
space RTP
Socket interface RTP payload

UDP
OS
IP UDP payload
Kernel
Ethernet IP payload

Ethernet payload
(a) (b)

Essence: RTP essentially just multiplexes a number

of multimedia streams into a single UDP stream. The
receiver is responsible for compensating missing pack-
ets (which is highly application dependent).

Real-time: RTP packets can be timestamped: pack-

ets belonging to the same substream can receive a
timestamp indicating how far off they are with respect
to their predecessor. This approach allows the system
to reduce jitter. In addition, timestamps can be used
to synchronize multiple substreams.
06 – 33 Transport Layer/6.4 UDP
 Transmission Control Protocol
(TCP)

• Connection-oriented service that supports byte

streams (not message streams). A sender may
send eight 512-byte packets that are received as
two chunks of 1024 bytes, and one of 2048 bytes.

• Transport address consists of a 16-bit port num-

ber, which augments the underlying IP address.

• TCP ensures reliable, point–to–point connections.

No support for multicasting or broadcasting.

• A TCP TPDU is called a segment, consisting of

(minimal) 20-byte header, and maximum total length
of 65,535 bytes. A segment is fragmented by the
network layer when it is larger than the network’s
maximum transfer unit (MTU).

06 – 34 Transport Layer/6.5 TCP

TCP Header
32 Bits

Source port Destination port

Sequence number

Acknowledgement number

TCP U A P R S F
header R C S S Y I Window size
length G K H T N N

Checksum Urgent pointer

Options (0 or more 32-bit words)

Data (optional)

• Acknowledgments are piggybacked when ACK =

1.
• SYN is for connection setup (ACK = 0: request;
ACK = 1: accepted).
• FIN is for connection release. Data sent before
the release is not lost.
• URG indicates immediate processing and trans-
mission: the receiver is signalled.

06 – 35 Transport Layer/6.5 TCP

 TCP Connection Management

Connection establishment: Uses three-way hand-

shake protocol.

Connection release: To be thought of as indepen-

dent releases of two simplex connections:
State Description
CLOSED No connection active or pending
LISTEN Server waiting for conn. request
SYN RCVD Conn. request has arrived; wait for ACK
SYN SENT Conn. request just sent; wait for SYN+ACK
ESTABLISHED Data can be sent and received
FIN WAIT 1 Client just sent conn. release
FIN WAIT 2 Server just agreed to release connection
TIMED WAIT Wait for all packets to die
CLOSING Client & server both tried to close
CLOSE WAIT Other side initiated release
LAST ACK Wait for all packets to die

06 – 36 Transport Layer/6.5 TCP

TCP Connection Management

Release
(Start)
CONNECT/SYN (Step 1 of the 3-way handshake)
CLOSED
CLOSE/–
LISTEN/– CLOSE/–
SYN/SYN + ACK
(Step 2 of the 3-way handshake) LISTEN

RST/– SEND/SYN
SYN SYN
RCVD SENT
SYN/SYN + ACK (simultaneous open)

(Data transfer state)

ACK/– SYN + ACK/ACK
ESTABLISHED
(Step 3 of the 3-way handshake)
CLOSE/FIN

CLOSE/FIN FIN/ACK
(Active close) (Passive close)

FIN/ACK
FIN CLOSE
CLOSING
WAIT 1 WAIT

ACK/– ACK/– CLOSE/FIN

FIN + ACK/ACK
FIN TIME LAST
WAIT 2 WAIT ACK
FIN/ACK

(Timeout/)

ACK/–
CLOSED

(Go back to start)

06 – 37 Transport Layer/6.5 TCP

 TCP Window Management (1/2)
Basic idea: The receiver sends an acknowledgment
for the next byte that can be sent in the current stream,
and the maximum number of bytes that may be sent.

Sender Receiver Receiver's

Application buffer
does a 2K 0 4K
write
Empty
2K S
EQ =
0

2K
ACK = 2048 WIN = 2048

Application
does a 2K
write 2K SEQ = 2048
Full

Sender is Application
IN = 0
blocked 4096 W reads 2K
ACK =
048
IN = 2
4096 W
ACK =
2K
Sender may
send up to 2K

1K
SEQ =
4096
1K 2K

06 – 38 Transport Layer/6.5 TCP

TCP Window Management (2/2)

Important: The TCP entity is not obliged to immedi-

ately transmit data that the application hands over: it
can do as much buffering as it likes. Same goes for
acknowledgments.

Example: Interactive, character-oriented applications.

Rather than sending one byte at a time, buffer as much
characters as possible until the previous batch is ac-
knowledged (Nagle’s algorithm). Note: we’re always
stuck to at least 40 bytes of overhead per TPDU.

Example: Avoid the silly window syndrome where

the server is reading one byte at a time (and acknowl-
edges one at a time). Instead, the receiver should wait
until it can receive a reasonable amount of bytes in a
row.

06 – 39 Transport Layer/6.5 TCP

 TCP Congestion Control (1/2)
Problem: As before, the transport layer has to take
into account that the underlying network can be the
bottleneck. Question is how to detect and react to
congestion.

Solution: use a congestion window next to the win-

dow granted by the receiver. The actual window size
is the minimum of the two.

• Initialize congestion window to maximum segment

size to be used in the connection. Send it off. If it
gets acknowledged, double the size. Repeat un-
til failure. Leads to initial congestion window size
(slow start).

• In addition, use a threshold. On a timeout, lower

the threshold to 50 % of the congestion window
size, do a slow start (exponential) until new thresh-
old, and add maximum segment size to conges-
tion window size after that (linear growth).

06 – 40 Transport Layer/6.5 TCP

TCP Congestion Control (2/2)

44
Timeout
40

36
Threshold
32
Congestion window (kilobytes)

28

24 Threshold

20

16

12

0
0 2 4 6 8 10 12 14 16 18 20 22 24
Transmission number

06 – 41 Transport Layer/6.5 TCP

 TCP Timer Management
Main issue: How do we determine the best timeout
value for retransmitting segments in the face of a large
standard deviation of round-trip delays:
0.3 T 0.3 T1 T2

0.2 0.2
Probability

Probability

0.1 0.1

0 0
0 10 20 30 40 50 0 10 20 30 40 50
Round-trip time (msec) Round-trip time (msec)
(a) (b)

RT T best current estimate of round-trip delay

D estimate of deviation of round-trip delays
M measured round-trip delay
RT T = αRT T + (1 − α)M
D = αD + (1 − α)|RT T − M|
timeout = RT T + 4 · D

06 – 42 Transport Layer/6.5 TCP

Wireless TCP
Problem: TCP assumes that IP is running across
wires. When packets are lost, TCP assumes this is
caused by congestion and slows down. In wireless
environments, packets get lost due reliability issues.
In those cases, TCP should do the opposite: try harder.

Solution #1: Split TCP connections to distinguished

wired/wireless IP:

Sender TCP #1 Base

station
TCP #2

Mobile
host
Router Antenna

Solution #2: Let the base station do at least some

retransmissions, but without informing the source. Ef-
fectively, the base station makes an attempt to im-
prove the reliability of IP as perceived by TCP.
06 – 43 Transport Layer/6.5 TCP