Transcript Lec 2 - University of Sydney

Network Architectures
Week 3 Part 2
Comparing The Internet & OSI
Internet “layers”
PDU





Application – layer 5
Transport – layer 4
Network – layer 3
Data Link – layer 2
Physical – layer 1
PDU (Protocol Data Unit)
Layers as per Kurose & Rose
Message
Segment
Datagram
Frame
Bit
Application – layer 5


Similar to OSI application layer
Examples of application services





E-mail – SMTP Simple Mail Transfer Protocol
FTP – File Transfer Protocol
Telnet – Remote login protocol
POP3 Post Office Protocol 3
HTTP – Hyper Text Transfer Protocol
The Web is not synonymous with the Internet – it is an
application service.
Transport – layer 4

Has two primary protocols


UDP – a connectionless “unreliable”
datagram service. Does not provide any retransmission or congestion control
TCP – provides a connection oriented
reliable data service with end-to-end error
detection and control
We will cover both of these in more detail later
Network – layer 3




This is the IP layer
Transfers packets from source to
destination via a connectionless
datagram service
Uses IP addresses as a road map to
locate a host within the Internet
Relies on routers to pass on the
datagram via the correct link
Data Link – layer 2


Moves data in frames from one node to
the next
Protocol may be different on each link
Physical – layer 1


Moves bit from one node to the next
Protocols relate to the different media
on each link
It is worth noting that the Data link and Physical layers are not
Specifically Internet layers – e.g. PPP, Frame Relay & ATM
Addressing within the Internet



We have talked about applications passing
messages through the network
We know it is more accurate to say that
Processes on each host pass messages
Thus for two processes to communicate we
need:


Source & destination host addresses specified by the IP
address
A processes on the hosts specified by a port number.
Common processes have specific post numbers


HTTP is port 80
SMTP is port 25
Domain Name Service - DNS


We prefer to use host names such as
“usyd.edu.au”
But the Internet uses a hierarchical IP address



In IPv4 this is a 32 bit address shown as a four byte address
eg 193.92.216.9, where each byte has a range 0-255 i.e.
9=00001001
In IPv6 this will become a 128 bit address, providing around
3 devices per square metre of the globe
Network Address Translation (NAT) where many numbers
internal to an organisation are not externally visible has
reduced the move to 128 bit addresses
DNS functions





DNS uses three levels of server to translate the
domain name into IP address –see Kurose for details
Hosts may have alias – DNS will find the real name
Load distribution – a domain may have several
replicated sites – DNS will provide the different IP
addresses to each successive request.
Application level protocol
Runs over UDP using port 53
Defining a Network
A Network within the Internet


A group of hosts isolated on one side of
a router
The path between two routers
Network IP addresses



Was defined in terms of A, B, & C classes
each with different numbers of devices
This was an inefficient method and was
replaced in 1993 with IETF’s Classless
Interdomain Routing (CIDR)
In this structure the network can be any
number of bits. It is expressed in the form
a.b.c.d/x where x is the number of bits
defining the network and 32-x is the number
of devices
Allocating Host addresses


Manually by the system administrator –
usually used for servers
Dynamically at boot time following the
Dynamic Host Configuration Protocol
(DHCP). This is normal for largish
organisations for workstations
Address Management



Domain names and IP Network addresses are
managed under the auspices of ICANN (The
Internet Corporation for Assigned Numbers &
Names)
Three registries: US, Europe, Asia Pacific
They resolve Domain name disputes and
maintain the DNS root servers
IPv6 is not universally accepted
Transport layer

Provides logical communication
between two processes on different
host. It provides two protocols:


UDP – User Datagram Service
TCP – Transmission Control Protocol
Transport layer

The transport layer is implemented on
each end host only
Application
Application
Transport
Transport
Network
Network
Network
Network
Link
Link
Link
Link
Physical
Physical
Physical
Physical
UDP


At the Transport layer it does very little. Passes the
message straight to the IP packet service
It is connectionless







No overhead in handshaking
No connection state in end systems
It has a smaller packet overhead
No congestion control
Trades off risk of data loss against higher throughput
Used in multimedia – telephony, video
Some apps. provide own error control
TCP – Reliable transmission service

To provide a reliable data transmission
service it needs:



An error detection mechanism. This is based on
a checksum calculation
A receiver feedback mechanism. The receiver
sends an ACK(nowledge) when an error free
packet is received, and sends a NAK when the
packet has errors.
A retransmission service is a packet is damaged
or not received
Three problems



How can you send a NAK if the packet
never arrives?
But if the second packet is not sent on
its way until an ACK is received back,
won’t it be slow? The “stop & wait”
problem
How do you prevent congestion in the
network?
Packets that do not arrive



Packets are given a sequence number and this number is
returned on the ACK or NAK. Thus the sending host knows which
packets have not been acknowledged and must be presumed lost
– it is an implied NAK
It is possible for packets to circulate forever and this problem is
prevented by giving the packet a life expressed in numbers of
links. This number is decremented at each router and the packet
discarded when the count is zero
The last problem is duplicate packets. The sequence number
resolves this one too.
Stop & wait problem


An example in Kurose shows an
effective transfer rate of 267kbps on a
1gbps line
The answer is pipelining or a sliding
window approach. This allows multiple
packets to be sent without waiting for
the acknowledgements. It can best be
shown in a diagram
Sliding Window Protocol
End-to-End flow control


It is relatively easily to send data too quickly, such that
the receiver or a router cannot handle the flow.
TCP provides congestion control





Host B establishes a connection buffer
Host B tells A how much room is available in the connection
buffer on each segment it sends back
Host A then has to ensure that the amount of data in the
pipeline, i.e. data that has not been acknowledged does not
exceed the connection buffer
It does attempt to dynamically increase flow rate and chokes
backs as congestion increases
This is a simplistic explanation – see Kurose page 246
Network layer

The network layer is implemented on
each host and router
Application
Application
Transport
Transport
Network
Network
Network
Network
Link
Link
Link
Link
Physical
Physical
Physical
Physical
IP




Provides a connectionless datagram service
Source to destination
Relies on routers to route the datagram
through the network
Routers use the IP addresses together with
their internal routing tables to direct
datagram down the appropriate link
IP Datagram header







IP protocol i.e. 4 & Transport protocol
Header length & header checksum
Datagram length – theoretically up to 64k but
rarely > 1,500 and often limited to 576
Message fragmentation information
Time to live
Source & destination IP addresses
Timing & routing parameters