The Transport Layer

Note on 22 Jul 2002
Services provided

• Goal of the transport layer is to provide efficient reliable & cost effective service to higher levels (normally applications)
• Normally provides connection oriented & connectionless services
• Functions in a similar manner to network layer

QoS

• Another way of looking at transport layer is that it's primary function is to enhance the "Quality of Service" of the network layer
• What exactly good is, is best left up to the user

  Connection establishment delay

  Connection establishment failure probability

  Throughput

  Transit delay

  Residual error ratio

  Protection

  Priority

  Resilience

Transport Service primitives

• Many programs & programmers use transport layer so primitives must be easy to use and understand

Primitive	PDPU sent	Meaning
LISTEN	none	Block until some process tries to connect
CONNECT	CONNECTION REQ.	Actively attempt
SEND	DATA	Send information
RECEIVE	none	Block until DATA TPDU arrives
DISCONNECT	DISCONNECTION REQ.	This side wants to release connection

Berkeley Socket primitives

BINDLISTEN

Primitive	Meaning
SOCKET	Create a new communication end point
	Attach a local address to a socket
	Announce willingness to accept connection: give queue size
ACCEPT	Block the caller until a connection attempt arrives
CONNECT	Actively attempt to establish a connection
SEND	Send data over the connection
RECEIVE	Receive data from the connection
CLOSE	Release the connection

Transport layer VS Data Link Layer

• Transport layer is similar in many respects to data link layer, but there are some important differences

	Datalink Layer	Transport Layer
Addressing	Unnecessary, output lines are already known	Explicit addressing is needed
Connection	Simple	Complex
Storage	Negligble	Significant
Amount of data	Continuous stream	Large, varying number of connections

Addressing

• When an application wishes to setup an connection - how to we specify which remote application to use?
  • Ans: Use Transport Service Access Point (TSAPs) and Network Service Access Points (NSAPs)
• How do we find out what TSAPs to use?
  • Well known services list
  • Directory server
• How do we find the NSAP for our TSAP?
  • Use a hierarchical TSAP address that contain NSAP
  • Use flat TSAP with name server to find NSAP
  • Ask

Establishing a connection

• Biggest problem with establishing a connection is the problem of duplicate packets
  1. Sender sends datagram to receiver, but due to congestion, these packets timeout
  2. Sender resends packets, which take a different route to receiver
  3. Old packets eventually get through congestion and reach receiver
• To solve problem need to restrict packet lifespan
  1. Restricted subnet design
  2. Hop counter
  3. Timestamping each packet

Establishing a connection - duplicate data packets

• let t = lifetime of a packet, let T = ct
• Equip each host with a time of day clock
  • that never stops
  • is a binary counter with number of bits >= number of bits in sequence number
• When connection is setup k low order bits are used as sequence number
• Sequence numbers are not allowed to be used for a time T before their use as initial sequence numbers

Establishing a connection - duplicate connection packets

Releasing a connection

• Asymmetric release
  • may loose data
• Symmetric release
  • Impossible to do correctly - 2 army problem
  • Timers must be used to help alleviate problem

Flow control & buffering

• Similar to data link layer, but transport layer has to deal with more connections
• If network service is unreliable then sender must buffer
• If network service is reliable and the receiver can guarantee to accept all TPDUs then the receiver can buffer, but the sender does not have to

Buffers

• The best place to buffer depends on the type of traffic carried
  • low bandwidth, bursty traffic - buffer at sender
  • high bandwidth, smooth traffic - buffer at receiver

Dynamic buffer allocation

No	From	Message	Comment
1	A->B	<req 8 buffers>	A wants 8 buffers
2	B<-A	<ack=15, buf=4>	B grants messages 0-3 only
3	A->B	<seq=0, data=m0>	A has 3 buffers left
4	A->B	<seq=1, data=m1>	A has 2 buffers left
5	A->...	<seq=2, data=m2>	Message lost, but A thinks it only has 1 left
6	B<-A	<ack=1, buf=3>	B acknowledge 0 & 1, permits 2-4
7	A->B	<seq=3, data=m3>	A has 1 buffer left
8	A->B	<seq=4, data=m4>	A has 0 buffers left and must stop
9	A->B	<seq=2, data=m2>	A times out and retransmit
10	B<-A	<ack=4, buf=0>	Everything acknowledge, but A still blocked
11	B<-A	<ack=4, buf=1>	A may now send 5
12	B<-A	<ack=4, buf=2>	B has found a spare buffer
13	A->B	<seq=5, data=m5>	A has 1 buffer left
14	A->B	<seq=6, data=m6>	A is now blocked again
15	B<-A	<ack=6m buf=0>	A is still blocked
16	..<-B	<ack=6, buf=4>	Problems!!

Crash recovery

• Problems can occur if host crashes during dta transfer

	Strategy used by receiver
	First Ack then Write			First write then Ack
	AC(W)	AWC	C(AW)	C(WA)	WAC	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

Internet Transport Protocols

• 2 main transport level protocols
• TCP - Transport control protocol
  • Guaranteed connection oriented service
• UDP - Universal Datagram Protocol
  • Unguaranteed connectionless service
  • Essentially IP with an additional header

TCP services

• TCP uses ports in range 0-65,535 (ports below 1024 are reserved and known as well known ports)
• TCP connections are full duplex and point to point
• TCP connections are bytes streams not message streams (i.e. message boundaries are not preserved by TCP)
• TCP is lazy - it may send all the data immediately, or it may buffer it to send later
• Users can use PUSH flag to force TCP to try and send immediately (like flush in file system)
• URGENT flag is also used to force TCP to send data immediately

TCP Protocol

• Basic protocol is variant on sliding window protocol
• Sequence numbers are 32 bits
• TCPs exchange data in the form of a segment, each segment consists of header plus zero or more data bytes
• Segment size is limited by
  • 65,535 IP limit
  • Maximum Transfer Unit (MTU) of network

TCP header

TCP connection management - making a connection

TCP connection management - breaking a connection

TCP transmission policy

TCP congestion control

• Standard TCP assumes that retransmissions are caused by congestion
• TCP assumes 2 types of congestion - each with their own window
  • Network congestion
  • Receiver congestion
• The number of bytes that can be transmitted at any one time is the minimum of the 2

Congestion control - Network congestion

• Slow start
  1. ms=maximum segment size in use on connection; cw=congestion window
  2. cw:= ms; n:=1
  3. send n segments & start timer
  4. if acknowledge then
     n:=2*n; cw:=n*ms
     goto 3
  5. if timeout then
     cw:=ms
• Internet congestion control algorithm
  1. threshold:=64k
  2. When timeout
     threshold:=1/2*cw; cw:= ms; n:=1
  3. Send n segments & start timer
  4. If acknowledge then
     if cw >= threshold then
     n:=n+1; cw:=n * ms; goto 3
     else
     n:=2*n; cw:= n* ms; goto 3
  5. If timeout then
     cw:=ms

UDP header