Computer Network Design - 6

(1)

Computer Network Design - 1 Andrea Bianco & Paolo Giaccone – TNG group - Politecnico di Torino

Review of fundamentals

networking concepts

Andrea Bianco Paolo Giaccone

Telecommunication Networks Group

[email protected]

http://www.telematica.polito.it/

Copyright

• Quest’opera è protetta dalla licenza Creative Commons NoDerivs-NonCommercial. Per vedere una copia di questa licenza, consultare

http://creativecommons.org/licenses/nd-nc/1.0/ oppure inviare una lettera a: Creative Commons, 559 Nathan Abbott Way, Stanford, California 94305, USA.

• This work is licensed under the Creative Commons NoDerivs-NonCommercial License. To view a copy of this license, visit:

http://creativecommons.org/licenses/nd-nc/1.0/

or send a letter to Creative Commons, 559 Nathan Abbott Way, Stanford, California 94305, USA.

Review

• What is a (modern) telecommunication and computer network? • Topologies

• Multiplexing/multiple access • Switching techniques • User (traffic) characterization

• Properties of telephone and computer networks • Protocol architectures for computer networks

– Physical layer (multiplexing) – Data link layer (multiple access) – Network layer (routing, congestion) – Transport Layer (congestion)

– Application Layer (client server, P2P, CDN, Data Center)

Networks

• Focus on modern telecommunication and computer

networks

– Telephone and Internet

• One possible network definition (service oriented

view)

– Infrastructure that provides services to applications:

• Web, VoIP, email, games, e-commerce, social nets, • phone calls, fax, …

– Internet

• provides programming interface to apps program to “connect” to Internet

• provides service options

Networks

• Another network definition:

– A set of nodes and channels that offer a connection among two or more points to make telecommunication possible among users (i.e. move infos (data/flows)

among nodes)

– Represented as a topology (graph)

• Different levels of detail

– Key issue in networks is resource sharing

• Node is the point where

– Multiplexing /multiple access (link sharing) and switching (node sharing) occurs

Type of channels

• Point-to-point channel

• Only two nodes connected to the channel

• The channel is used by both nodes (often in

the same fashion)

• Sharing the channel among flows is called

multiplexing

(2)

Type of channels

• Broadcast channel

– Many Tx and many Rx

• Sometimes one (many) Tx and many (one) Rx

• Single communication channel shared by all nodes

– This room!

• The information sent by one node is received by all

other nodes (with some delay)

• Sharing the channel among flows is called multiple

access

Channel properties

• Many quality indices

– Attenuation, robusteness to mechanical stress, ease of installation, robusteness to interference, cost, etc,

• Mainly interested in

– bit rate [bit/s]

• Also named bandwidth, throughput, with slightly different meaning

– delay [s]

• propagation delay • depends on the channel length

– Bandwidth x delay [bit]

• channel ‘’size’’

• how much we can ‘’push’’ on the channel

Topologies in TLC networks

• The network topology is defined by the relative position of nodes and channels

• A network topology is a graph G=(V,A)

– V = set of vertices (represented as circles - nodes) – A = set of edges (represented as segments - channels)

• Edges may be:

– direct (directed segments (arrow) – unidirectional channels) – undirect (non directed segments – bidirectional channels)

• Abstraction of real networks

– Several levels of abstraction are possible

Many topologies

• Full mesh, mesh, tree, bus, star, …

A E B C D A E B C D A E B C D A E B C D A E B C D A E C B D A E B C D

Physical and logical topology

• It is important to distinguish between the

physical and the logical topology

– Logical topology: logical interconnection among

nodes via logical channels

– Physical topology: takes into account transmission

media constraints

Physical topology

• E.g., satellite network

A B

(3)

Logical topology

• Private network over satellite network

A B

C D

Logical topology

• Router interconnection for an ISP

Frame Relay or ATM network Router Router _Router Router Node Node Node Node Virtual Circuits

Logical topology

• Overlay among peers in a P2P network

Logical topology

access net access net access net access net access net access net access net access net access net access net access net access net access net access net access net access net ISP B ISP A ISP C IXP IXP regional net

Logical topology

access ISP access ISP access ISP access ISP access ISP access ISP access ISP access ISP Regional ISP Regional ISP

IXP IXP

Tier 1 ISP Tier 1 ISP Google

IXP

(4)

Topologies and performance

• The amount of traffic that can be succesfully

transferred (throughput) in a network is

– for a given channel available capacity

– inversely proportional to the average distance

among node pairs

– weighted by the amount of traffic exchanged

between the two node

• For uniform traffic and regular topologies the

average distance on the topology establish the

throughput

Topologies and performance

• Comparison among topologies, with the same

number of nodes (4) and (almost) the same

number of channels

• Uniform traffic

– Every node pair exchange x bit/s. Total generated

traffic is 12x.

• Every unidirectional channel has capacity B bit/s.

• Compute: average distance, network capacity

(maximum throughput), maximum channel

load,maximum node load

Topologies and performance

• Capacity: 3x2B=6B

• Average distance: 20/12=1.66

• Consider only traffic from left to right (simmetry)

– maximum channel load is 4x. Thus, x <= B/4

• Node 3 (or 2) must handle 7B/4 of traffic unit

• Uniform traffic, non regular topology, unbalanced

channel load, unbalanced node load

1

2

3

4

3x 2x 2x x x x

Topologies and performance

• Capacity: 3x2B=6B • Average distance: 1.5

• Considering the traffic from node 4.

– Maximum load on (all) channel is 3x. Thus x <=B/3 – The same holds for the other direction

• Node 4 must handle 3B of traffic unit

• Uniform traffic, non regular topology, balanced channel load, unbalanced node load

x x x x x x x x x 1 2 4 3

Topologies and performance

• Capacity: 4x2B=8B • Average distance: 1.33

• For clockwise traffic the maximum channel load is 2x. Thus x <= B/2.

– The same holds for counter clock wise traffic

• Each node must handle 2B unit of traffic

• Uniform traffic, regular topology, balanced channel load, balanced node load 3x/2 x/2 3x/2 x/2 3x/2 x/2 3x/2 x/2

Channel, sharing

(5)

Sharing channel resources

• Sharing of channel resources among data flows comes in two different flavours

– Multiplexing

• All flows access the channel from a single point • Single transmitter scenario

• Centralized problem

• A radio access from an antenna (base station in a cellular network, access point in a WI-FI network, satellite transmission), an output link in a switch or a router

– Multiple-access

• Flows access the channel from different access points • Many transmitters are active

• Distributed problems

• Local area networks (if not switched), mobile phones in a cellulare network, PC accessing via a Wi-FI hot-spot

Channel sharing techniques

• Frequency (FDM - FDMA)

• Time (TDM - TDMA)

• Code (CDM - CDMA)

t

f

channel

Frequency division

• Each flow is transmitted using different frequency

bands

• Need for band guard

f

t

Time division (TDM – TDMA)

• Each flow exploits different time intervals (slots)

• Define frame over which slot allocations are repeated

• Need for time guard

f

t

Code division

(CDM – CDMA)

• Each flow exploits a different code (waveform

with higher frequency than the bit tx rate)

• Need for orthogonal codes

c

t

f

Code division

(CDM – CDMA)

f

(6)

Code division

• Example

– Code word used by user i: +1 +1 -1 -1

– Coded sequence = information bit x code word

– Information bit:

-1

-1 1 1 -1

– Coded sequence:

-1-1+1+1 -1-1+1+1 +1+1-1-1 +1+1-1-1 -1-1+1+1

Code multiplexing

• Example:

– Code word for user 1: +1 +1 -1 -1

– Code word for user 2: +1 +1 +1 +1

– Code word for user 3: +1 -1 +1 -1

– Code word for user 4: +1 -1 -1 +1

• We are interested in receiving data from user 1

• Over the channel, transmitted signals sum up

(need to equalize power)

– Tx of 1+2+3:

+3 +1

+1

-1

– Tx of 2+3 (noise):

+2 0 +2 0

Code multiplexing

• Reception = correlation with code words

– Reception of user 1 = scalar product of the received

sequence with the code word +1 +1 -1 -1

– Everything we receive

• Transmissions of 1+2+3: +3 +1 +1 -1 • Correlation with +1 +1 - 1 -1 = 4

– The noise

• Transmissions of 2+3: +2 0 +2 0 • Correlation with +1 +1 -1 -1 = 0

Multiplexing or multiple access

• Time, frequency, code (and space: multiple

parallel wires) are all equivalente alternatives

– Given a channel capacity we can choose one among

the above techniques depending on technological

constraints

• Code division permits to “increase” channel

capacity (by allowing more users) if using

pseudo-orthogonal codes but degrading the

signal to noise ratio at the receiver (increase the

bit error rate)

FDM and TDM

FDM frequency time TDM frequency time 4 users Example: 125 ms

Statistical multiplexing

• Multiplexing can be

– deterministic, fixed in time, on the basis of

requirements determined at connection setup

– statistical, variable in time, to adapt to instantaneous

(7)

Statistical Multiplexing

• Sequence of A & B packets does not have fixed

pattern, bandwidth shared on demand

• Dynamic TDM scheme

A B C 100 Mb/s Ethernet 1.5 Mb/s D E statistical multiplexing queue of packets waiting for output

link

Switching techniques

Circuit and packet switching

Switching techniques

• Circuit switching

– Telephone networks

• Packet switching

– Two flavours (datagram and virtual circuit service)

– Internet, computer networks

Circuit switching

• Three phases

– Opening – Data transfer – Closing

• Network transparently

forwards user

generated data

– No operation on user data

• Fixed bit rate 64kbit/s

U1 N1 N2 U2

t t t t

Space vs time switching

125 ms

U1 N1 N2 U2

Switching techniques

• Circuit switching

– Resources allocated uniquely to a circuit

• Physical channel, time-slot in TDM frame

• Resources are shared among connections but are statically allocated to data

– Connection oriented

• Need to open (and close) the circuit prior (after) data transmission • Store state information on each circuit (stateful approach)

– Address (unique for each user in the network) used only when opening the circuit, not carried in data

– Data unit identified by position

– Routing (choice of the best route) performed only when opening the circuit

• Done through routing table lookup

– Data forwarding

• Through forwarding table look-up (one entry for each active circuit) • Static (always the same scheduling, unless circuits are closed or

(8)

Packet switching

U1 N1 N2 U2 t t t t Transmission time Propagation time Processing time

Data transfer over virtual circuits

Forwarding table DTE1 1 2 DTE2 3 4 5 1 2 3 4 5 1 2 3 1 2 3 4 5 32 612 1 32 5 612 216 314

In Label Out Label

Grouping virtual circuits

• A virtual circuit is logically identified by a

label

– The label may change value over each link

• Relabeling

• Label = often a pair of identifiers (VCI-VPI in

ATM)

– Virtual channel (VC): identifies a single

connection

– Virtual path (VP): identifies a group of virtual

channels

Grouping virtual circuits

• The grouping allows flow aggregation

– Eases network management

– Increases scalability

• Possible use

– LAN inter-connection to crete a VPN (Virtual

Private Network)

– Multimedia flows (video, audio, data)

VPI 1

VPI 6

VCI 1 VCI 2 VCI 3 VCI 4 VCI 5

Virtual circuits and paths (ATM)

VP SWITCHING VCI 21 VCI 22 VCI 23 VCI 24 VCI 25 VCI 24 VCI 23 VCI 24 VCI 25 VCI 24 VCI 21 VCI 22 VPI 1 VPI 2 VPI 3 VPI 8 VPI 4 VPI 5

ATM: VP switching

(9)

VCI 25 VCI 25 VCI 21 VCI 21 VCI 23 VCI 24 VCI 23 VCI 24 VC SWITCHING VPI 4 VPI 5 VPI 2

ATM: VC switching

Virtual circuits

• Switched virtual circuit (SVC)

– Established on-demand, through signaling, in real-time – Three phases

• Virtual circuit opening • Data transfer • Virtual circuit closing

– Users (and network) exchange signaling packets (over dedicated VCI/VPI) to establish a virtual circuit; then, data transfer can occur

• Permanent virtual circuit (PVC)

– Established through agreement among user and network provider

• Off-line, through management procedures

– Define a semi-static network

• Logical topology

– Users can immediately exchange data, with no delay

Switching techniques

• Packet switching, with datagram service

– Shared resources

• Ideally the full network is available to a single user

• Resources are “continuously” shared with all other users at the data level

– Connectionless

• Free to send data when available, no need to check network or user availability

• Stateless approach

– Each packet must carry the destination (and source) address – Data unit identified through source and destination addresses

(unique for each pair of users in the network)

– Routing and forwarding performed independently over each packet

• Through routing table look-up

Switching techniques

• Packet switching, with virtual circuit service

– Shared resources

• Resources are shared with all virtual circuits sharing the same link – Connection oriented

• Need to open (and close) the virtual circuit prior (after) data transmission

– Permanent virtual circuits available

• Store state information on each virtual circuit (stateful approach) – Address (unique for each user in the network) used only when opening the

virtual circuit, not carried in data

– Data unit identified through a label (unique for each existing virtual circuit on each link in the network)

• Label is unique on each link, but has a local scope, i.e. the value assumed is different on each link for simplicity

– Routing (choice of the best route) performed only when opening the virtual circuit

• Done through routing table lookup – Data forwarding

• Through forwarding table look-up (one entry for each active virtual circuit) • Re-labelling needed

Fundamentals of packet

switching

• Data sent as packets

• Nodes operate in store&forward (almost always)

– Buffers – Delay

• Many operations on data in the network (not in

circuit switching)

– Error detection, error recovery, flow control, routing, forwarding, congestion control, packet inspection …….. – Need to define a network architecture to organize

functionalities (see later)

Packet size

• Packet size P

– Measured in bit

• Packet size in time T

TX

– Transmission time measured in s

– Different on every link

– T

TX

= P/V

TX

where V

TX

is the link bit rate

• Packet size in meter M on a given link

(10)

Delays

• Delays suffered by each packet from source to destination node

• In each link

– Transmission (and reception) delay

• It is a funciton of the packet size in bit and of the link bit rate

– Propagation delay

• It is a function of the link length in meters

• In each switching node

– Processing time

• Function of the processing speed and of the complexity of the executed procedures on the packet header

• Normally negligible with respect to transmission time

– Queuing delays

• Depend on the traffic generated by all users • Highly variable

Traffic characterization

and QoS Provisioning

User traffic characterization

• Need to know user behaviour to design a

network

• Traffic sources

– CBR (Constant Bit Rate)

– VBR (Variable Bit Rate)

• Characterized by their rate [bit/s]

User traffic characterization

• CBR (Constant Bit Rate) sources:

– Rate (bit/s)

• “Perfectly” known

– Call duration (s)

– Call generation process

• Only statistically known

User traffic characterization

• VBR sources:

– Average rate (bit/s)

• Known? • Over which period?

– Peak rate (bit/s) or

– Burstiness (Peak rate/ average rate)

• Known (worst case)

– Burst duration

• Known?

– Call duration (s) – Call generation process

• Only statistically known

low speed data alphanumeric terminals LAN alta velocità immagini connectionless data transmission supercomputer interconnection voice audio HDTV VIDEO compressed non compressed B urs ti nes s

Peak rate [bit/s]

103₁₀4 ₁₀5 ₁₀6 ₁₀7 ₁₀8 ₁₀9 ₁₀10 1000 100 10 1 LAN graphical terminal

Burstiness= Peak rate/ Average rate

(11)

Integrated vs dedicated networks

• Telecommunications networks were traditionally

defined to provide a specific service

– one service one network paradigm

• Telephone network for the interactive human voice transportation service

• Internet for data exchange among computers • TV or radio distribution for the TV or radio system

• Integrated networks

– one network for any service

• narrowband ISDN o N-ISDN • broadband ISDN o B-ISDN

Integrated vs dedicated networks

• Dedicated networks

– Easier to optimize for the specific service

– “Optimal” engineering solutions for the specific requirements of the service

• Integrated networks advantages

– No need to create an independent infrastructure for each service

– Supporting different requirements implies sub/optimal choices

• Integrated networks trade flexibility and infrastructure

cost reduction with perfomance and increased control

complexity

Quality of service provided by

networks

• Networks used as examples

– Fixed telephone network: POTS

– Internet

– B-ISDN

• Describe in an informal way the quality of

service provided by these networks

POTS

• Characteristics

– CBR source completely known (generated by the network)

– Circuit switching

• Constant, dedicated bit rate  no congestion

• Minimum possible delay (only propagation): order of tens of ms (real time)

• Zero loss probability

• Requirements

– Error probability smaller than few % – Small or negligible blocking probability

• QoS largely independent of other users (apart from

blocking probability)

• Network utilization can be really low, user

satisfaction very high

Internet

• Characteristics

– Source behavior unknown

– Packet switching with datagram service

• Complete sharing of network resources • Bit rate and delay unknown • Possible congestion

• Loss probability may be significant

• Requirements

– Error probability negligible in wired networks

• Zero blocking probability

• QoS largely dependent of other users

• Network utilization can be very high, user

satisfaction can be very low

B-ISDN

• Intermediate situation

– Source known (either deterministically or statistically) – Packet switching with virtual circuit service

• May introduce algorithms to control network resources sharing • Bit rate and delay negotiable

• Loss probability negotiable

– Requirements

• Blocking probability reasonably small • Error probability negligible

• QoS dependent of other user behavior and of

algorithms used to manage network resources

• Trade network utilization and user satisfaction

(12)

Network design

• Design problem

– Given:

• Network topology (nodes, link speed) • Traffic characterization

– Jointly obtain:

• Guarantee some sort of QoS for user connection • High network utilization

• Without the objective of high network utilization, the

problem becomes “trivial”

– overprovisioning (power line or water distribution networks)

Network design

• Several flavour

• Running at different time scale (with different level of complexity)

• Mainly focus on network design and planning (resource deployment)

• On the basis of traffic estimates and cost constraints • Exploits routing criteria and traffic engineering

• Other examples

– Network management (running a network)

• Measurements

• Fault management (protection and restoration) • Includes re-design and re-planning

• Connection management • Data unit transport

Design to obtain QoS

• Different time scale (with different level of

complexity)

• Network design and planning (resource

deployment)

– On the basis of traffic estimates and cost constraints – Exploits routing criteria and traffic engineering

• Network management (running a network)

– Measurements

– Fault management (protection and restoration) – Includes re-design and re-planning

• Connection management

• Data unit transport

Our definition of QoS

• Assume that a network has been designed and is properly managed

– Available resources are given

• Mainly study algorithms operating at the following time-scale:

– Connection management – Data unit transport

• Also named traffic control problem

• Must define what is meant by connection. Also named data classification problem.

• Two different traffic control principles:

– Preventive control : mainly executed at network ingress, with fairly tight traffic control to avoid congestion insurgence in the network – Reactive control: react when congestion situation occur, to reduce or

eliminate congestion negative effects

Traffic control:

essential elements

• Connection oriented network

• User-network service interface

– Traffic characterization – QoS negotiation

• Resource allocation (bit rate and buffer)

• Algorithms for traffic control

– CAC (Connection Admission Control) and routing – Scheduling and buffer management (allocation, discard)

in switching nodes

– Conformance verification (policing or UPC: Usage Parameter Control)

– Traffic shaping to adapt it to a given model – Congestion control

Traffic control:

connection oriented network

• The connection oriented paradigm permits to know

which are the network elements over which traffic

control algorithms must be executed (path known)

– Circuit switching

– Packet switching with virtual circuit service

• If high utilization is a major objective:

– Packet switching

• As such, the most suited switching technique to

obtain QOS is packet switching with virtual circuit

service

(13)

user-network service interface

• The capability to control the network increases with the knowledge of user traffic. Limiting factor is the complexity. • Over the service interface

– Traffic characterization – QoS parameters negotiation

• Can be defined on a call basis or on a contract basis • POTS: implicit, on a contract basis

• Internet: not existing

• Frame relay: negotiable, normally on a contract basis • B-ISDN: negotiable with traffic contract on both contract and

call basis

• Internet extended to support QoS: negotiable through a SLA (Service Level Agreement) mainly on a contract basis

resource allocation

• Main resources:

– Bit rate over transmission links – Buffer

• Resources can be allocated

– On a contract basis (booking) – On a call basis

– Packet by packet

• Allocation

– Exclusive (dedicated resource) – Shared

Algorithms: CAC and routing

• Routing

– QoS based path selection to router a connection

• CAC

– Determine whether to accept a connection or not, depending on

• The path chosen by the routing algorithm • Traffic characterization

• QoS requests • Network status

• Constraints

– It is not acceptable to destroy or even reduce the quality of service guaranteed to already accepted connections 

• Can be relinquished

– Connection must be refused to avoid network overload or congestion

• Preventive control (but can become reactive)

Algorithms: scheduling and

buffer management

• Scheduling

– Choice of the data unit to be transmitted among data unit stored in the switch

• Buffer management

– Allocation (partial/total, exclusive/shared) of memories in the switch

– Dropping policies

• Mandatory in an heterogeneous environment to support different QOS requests

– FIFO (First In First Out) or FCFS (First Came First Served) policy with drop-tail discard is optimal in a homogeneous environment – Counter for less than 10 pieces at supermarket

• Preventive and reactive

Algorithms: policing e shaping

• Policing (traffic verification)

– Network control of user behavior to guarantee conformance to traffic characterization

• Shaping (traffic conditioning)

– User/network adaptation of data traffic to make it conformant to a given characterization

• Mandatory to control user honesty and to adapt traffic which is difficult to generate as conformant a priori

• Where algorithms must be executed?

– Only at network edge, i.e., when user access network? – Multiplexing point modify traffic shape

• Both at network access and internally to the network

• Mainly preventive, but they can become reactive if QoS level may change over time

Algorithms: congestion control

• Congestion

– Traffic excess over a given channel (link)

• Can occur due to

– Short term traffic variability

– Allocation policies that share resources to increase network utilization

• Congestion effects:

– Buffer occupancy increase – Delay increase

– Data loss

• Needed to obtain high link utilization

• Must execute at network edge, within the network

or….?

(14)

Protocol architectures

Architectures and protocols

• Communication requires cooperation

• One abstract description of the communication

paradigm between two or more users requires

the definition of a reference model specifying a

network architecture

• A network architecture defines

– the communication process

– the relation among objects used in

communication

– the functionalities to support the communication

• Layered architectures!

Layered architectures

• Layered architectures are used because of

– simple design – simple management – simple standardization – separation among functions

OSI Application Presentation Session Transport Network Data Link Physical DECNET User Netw. Appl. Session End to End Routing Data Link Physical ARPA Application Service Internetwork Network SNA Transaction Service Presentation Service Data Flow Trans. Control Manag. Service Virtual Route Explicit Route Transm. Group Data Link Physical path control half session

router 1 router 2 router 3 host 1 host 2 host 3 host 4 subnet 1 subnet 2 subnet 4 subnet 3 packet transfer routing error control applications

Separation among functions:

Internet

Management plane

Control plane User plane

High layers AAL ATM Physical High layers

B - ISDN

Protocols

• Formal definition of the procedures adopted to

guarantee the communication between two or

more objects on the same hierarchical level to

execute a specific function

• Protocol definition:

– semantics

• set of commands and answers

– syntax

• structure of commands and answers

– timing

(15)

Layer and entities

• Active elements in a subsystem

• Run the functions of the layer (exploiting service provided by lower layers)

• Interact (cooperate) within the same layer

(N) - layer

System A System B

(N) - entity

transmission media

Services

• A service can be:

– connection-oriented (CO): a preliminary agreement

(connection) is established between the network and

the communication end-points, then the data is

transferred and finally the connection is released

– connectionless (CL): data is sent to the network

without any preliminary agreement and is treated

independently from each other

(N) - service N+1 N N+1 N (N) – service provider

Black-box for the (N+1) - entity

Services

N N-1 N N-1 (N-1) - service (N-1) – service provider

Black-box for the (N) - entity

Services

(N) - layer interface (N-1) - layer (N-1) - SDU (N-1) - PCI (N-1) - SDU (N-1) - PDU

SAP

(N) - PDU

PDU creation

Application Presentation Session Transport Network Data link Physical Receiver Transmitter Application Presentation Session Transport Network Data link Physical data APCI ASDU PPCI PSDU SPCI SSDU TPCI TSDU NPCI NSDU DLPCI DLSDU bit or symbols

Information transfer

(16)

application presentation session transport network data link physical transmission media Application protocol Presentation protocol Session protocol Transport protocol Network protocol Data link protocol Physical layer protocol

application presentation session transport network data link physical

The seven OSI layers

Functions (OSI-like)

• Physical layer:

– allows to transfer binary digits exchanged among the data link entities

– deals with bits or symbols

– defines transmission codes, connectors, voltage levels, etc…

• Data link layer

– error detection and error correction – flow control

– data unit (packet, cell, datagram) delimitation – addressing

Functions

• Network layer

– routing

– congestion control (moved to layer 4 in the Internet) – pricing

– addressing

• Transport layer

– error control – sequence control – flow control (end to end) – congestion control (Internet)