Internet architecture & addressing

6CCS3INS Internet Systems

2014-15 Toktam Mahmoodi, Department of Informatics, KCL
Questions

 How can we divide up the administration of the internet

to make it manageable?

 When one part is changes, how can we avoid replacing

all the software which runs the internet (maintenance)?

 How can a host identify another host to send data to,

when they are not directly connected?

 How can our addressing schemes handle varying

numbers of hosts in organisations?
Outline

 Internet architecture

 Addressing hosts on the internet

 IPv4 addressing
 IPv6 addressing

 Routing Protocols
Internet Architecture

 The main goal of the Internet Architecture was to:

 connect existing networks
 be robust in case of different link failures or whole sub-networks
 routing functionality that can adapt to this phenomena
 support multiple types of services
 allow distributed management
 allow host attachment with a low level of effort
 be cost effective
 In terms of header overhead, retransmissions and routers
capabilities needed
Internet Architecture: Layers

 Many Internet communication issues, including

 Routing, Reliability, Flow control, Data format

 Internet as stack of layers each dealing with one or a few issues

 Different ways to solve each issue, so different implementations of each layer
 Combine the best solutions for any one problem

 Protocol Stack:
 Physical layer provides the actual connectivity (copper, fiber, radio)
 Access layer defines how to deliver data between
Application
two devices on the same network
 Network layer defines how to route messages
Transport
across networks
Network
 Transport defines how to provide reliable communication,
so that data will not be lost or corrupted (TCP, UDP)
Access/Link
 Application layer defines how programs request messages
to be sent across the internet, using encryption, compression etc. Physical
The Internet Protocol

 IP is the network layer, a packet delivery service (host-to-

host), that is
 Connectionless: each datagram is independent of all others.
 Unreliable: there is no guaranteed delivery

 Responsible for:
 Fragmentation / Reassembly (based on MTU).
 Routing.

 Internet Control Message Protocol (ICMP)

 Part of and supports IP
 Used to report errors and other info back to sender of IP message
 Itself sent over IP
Internet Architecture: edge-oriented

 The success of the Internet is due to the edge-oriented

approach of the architecture.
 A connectionless, packet-forwarding infrastructure (“dumb network”)
that positioned higher-level functionality at the edge of the network for
robustness.

 The end-to-end argument, intelligent edges and dump

network or in other words keep the infrastructure as simple as
possible.
 To reduced complexity of the core network.
 New applications can be added without changing the core (look for
example in p2p!)
 This is known as the hour glass model (with IP at the waist).

 With addresses that are

 fixed-size numerical quantities, with a simple (net, host) hierarchy
 applied to physical network interfaces, which can therefore be used for
both naming a node and for routing to it.
 Packet Transmission

HTTP
MESSAGE

A B

HTTP
User A runs HTTP application TCP

IP

Link Layer

Copper
 Packet Transmission

TCP HTTP
Header MESSAGE

A B

HTTP

TCP adds header including packet TCP

number, and the time out
IP

Link Layer

Copper
 Packet Transmission

IP TCP HTTP
Header Header MESSAGE

A B

HTTP
IP adds header including host A and B TCP
addresses
IP

Link Layer

Copper
IP Fragmentation

 Different link layer technologies can carry packets of different size.

 The maximum amount of data that the a link layer packet can carry is called
Maximum Transfer Unit (MTU).
 Because the IP is encapsulated within the link layer this MTU puts a limit in
the maximum size of the IP packet.

 If the outbound link has smaller MTU than the IP packet that the
router want to sent, the solution is Fragmentation
IP Header

 IHL: header length only in 32-bit words (5 <= HLen <= 15)
 Type of Service, split in:
 Differentiated Service Field (6 bits)
 remaining two bits used by ECN (Early Congestion Notification)
 Length: the length of the entire datagram/segment; header + data
 Flags: Don’t Fragment (DF) and More Fragments (MF)
 Fragment offset: all fragments excepting last one contain multiples of 8 bytes
The Internet Engineering Task Force (IETF)
produce technical documents that influence the way people design, use,
and manage the Internet – RFCs (Request for Comments)
Outline

 Internet architecture

 Addressing hosts on the internet

 IPv4 addressing
 IPv6 addressing

 Routing Protocols
Hierarchical addressing

Internet
A
Q
D 1

Address of Host

A- D - Q- 1
IP addresses

 IP includes a structure for addressing hosts

 In the global Internet every host and router must have one globally unique
address
 IP address technically are associated with an interface and NOT with a host

 Currently, the internet primarily uses version 4 (IPv4)

 IPv6 compatibility is being deployed slowly.

 Why IPv6 is needed (short/long term advantages)

 Increased Address Space
 We are running out of Internet addressing.
 Lack of Security at the Network layer
 Quality of Service for end-to-end networking.
 supporting new features for applications
Outline

 Internet architecture

 Addressing hosts on the internet

 IPv4 addressing
 IPv6 addressing

 Routing Protocols
IPv4 Addressing

 IPv4 addresses are 32bits long.

 With N bits, we can allocate 2N different addresses
 32 bits = 4,294,967,296 different addresses

 In order to provide the flexibility required to support different size

networks, the designers decided to divide the IP address space into
three different address classes:
 Class A, Class B, and Class C - “classful” addressing.
 Two additional classes later added:
 Class D addresses used to support IP Multicasting.
 Class E addresses are reserved for experimental use.

 Every IP address consists of two parts, one identifying the network and
one identifying the Host.
 The Class of the address and the subnet mask determine which part belongs to
the network address and which part belongs to the node address.
IPv4 addresses by class

 IP addresses were split into

 Class ID
Network prefix
 Network ID
 Host ID

 Class A 0

 Class B 10

 Class C 110
Class A addressing (or /8s networks)

 Class A is for very large networks

 8-bit network-prefix with the highest order bit set to 0
 24-bit host-number

 A maximum of 126 (2^ 7 -2) /8 networks can be defined.

 the /8 network 0.0.0.0 is reserved for use as the default route and the /8 network 127.0.0.0
(also written 127/8 or 127.0.0.0/8) has been reserved for the "loopback" function.

 Each /8 supports a maximum of 16,777,214 (2^24 -2) hosts per network.

 all-0s ("this network") and all-1s ("broadcast") host-numbers may not be assigned to
individual hosts.

 Since the /8 address block contains 2^31 individual addresses and theIPv4
address space contains a maximum of 2^32 (4,294,967,296) addresses,
 the /8 address space is 50% of the total IPv4 unicast address space.
 range: from 1 to 126
Class B addressing (or /16s networks)

 Class B is for much smaller networks

 16-bit network-prefix with the two higher order bits set to “10”
 16-bit host-number

10

 A maximum of 16,384 (2^ 14 ) /16 networks can be defined

 Each /16s supports 65,534 (2^16 -2) hosts per network.

 Since the entire /16 address block contains 2^30 (1,073,741,824)

addresses,
 it represents 25% of the total IPv4 unicast address space.
 Range from 128 to 191
Class C addressing (or /24s networks)

 Class C is for much smaller networks

 24-bit network-prefix with the three higher order bits set to “110”
 8-bit host-number

110

 A maximum of 2,097,152 (2^ 21 ) /24 networks can be defined.

 Each /24s supports 254 (2 ^8 -2) hosts per network.

 Since the entire /24 address block contains 2^ 29 addresses,

 it represents 12.5% (or 1/8th) of the total IPv4 unicast address space.
 Range from192 to 223
Classful addressing

 Limitations with the classful addressing:

 A /24, which supports 254 hosts, is too small while a /16, which supports 65,534 hosts, is too large.
 For example an organization with 2000 hosts will go for /16!.

 1993 IETF standardised the Classless Interdomain Routing (CIDR)

 In CIDRised networks the Network part of the IP address can be ANY number of bits long (rather
than being constrained to 8,16 or 24)

 For the example of organisation with 2000 hosts,

 IP addresses of the form a.b.c.d/21 can be allocated (2046 hosts).
 the first 21 bits specify the organisation’s network address and are common for all the hosts inside
the network,
 The remaining 11 bits can specify a specific host inside this network.
 In a real case scenario the 2000 hosts will be further divided using the last 11 bits in subnets to
create different networks inside the organisation

 Today, address classes are ignored.

 Instead, routers are explicitly told the length of prefix to look for
Example IP addresses

 1.22.11.12 Class A
 137.22.11.12 Class B
 201.22.11.12 Class C
 193.92.96.255 /24 Broadcast address
 All “1”s at the host part represent the broadcast address.
 The broadcast address is the address used to send a message
to all hosts on the network
 193.92.96.0 /20 Network address
 All “0”s at the host part represent the network address
 The network address is an IP address for the network as a
whole, as used by routers to know where to send the messages
Subnets

 RFC 950 defined a standard procedure to support subnetting

of a single Class A,B or C network into smaller pieces.
 Internet Routing Tables were beginning to grow
 Local administrators had to request another network number from
the Internet before a new network installed into the organization.

 These problems were attacked by introducing another level of

hierarchy to the IP addressing structure
Network Mask

 In a Binary format the 1’s represent the portion of the IP

address that is Network number and the 0’s represent
the portion of the IP that is the Host number

 Default Subnet Masks

 Class A:
11111111.00000000.00000000.00000000 255.0.0.0
 Class B:

11111111.11111111. 00000000. 00000000 255.255.0.0

 Class C:

11111111.11111111.11111111.00000000 255.255.255.0
Example 1 – part 1

 Organisation has assigned the network number 193.1.1.0/24

and wants to have 6 subnets, where the largest one need to
support up to 25 hosts.

 The first step is to determine the number of bits required

to define the six subnets.
Example 1 – part 2

 Organisation has assigned the network number 193.1.1.0/24

and wants to have 6 subnets, where the largest one need to
support up to 25 hosts.

 We have two unused subnets that can be reserved for future

growth.

 A 27-bit ENP leaves 5 bits to define host addresses on each

subnet.
 each subnetwork with a 27-bit prefix represents 32 individual IP
addresses.
 since the all-0s and all-1s host addresses cannot be allocated, there
are 30 assignable host addresses on each subnet.
Example 1 – part 3

 Organisation has assigned the network number 193.1.1.0/24

and wants to have 6 subnets, where the largest one need to
support up to 25 hosts.
 The eight subnets then will be:
Base Net: 11000001.00000001.00000001 .00000000 = 193.1.1.0/24
Subnet #0: 11000001.00000001.00000001. 00000000 = 193.1.1.0/27
Subnet #1: 11000001.00000001.00000001. 00100000 = 193.1.1.32/27
Subnet #2: 11000001.00000001.00000001. 01000000 = 193.1.1.64/27
Subnet #3: 11000001.00000001.00000001. 01100000 = 193.1.1.96/27
Subnet #4: 11000001.00000001.00000001. 10000000 = 193.1.1.128/27
Subnet #5: 11000001.00000001.00000001. 10100000 = 193.1.1.160/27
Subnet #6: 11000001.00000001.00000001. 11000000 = 193.1.1.192/27
Subnet #7: 11000001.00000001.00000001. 11100000 = 193.1.1.224/27
Example 1 – part 4

 Organisation has assigned the network number 193.1.1.0/24

and wants to have 6 subnets, where the largest one need to
support up to 25 hosts.
 For the host addresses we are using all the 5 bits except all 0
(subnet) and 1’s (broadcast)
Subnet #2: 11000001.00000001.00000001.01000000 = 193.1.1.64/27
Host #1: 11000001.00000001.00000001.01000001 = 193.1.1.65/27
Host #2: 11000001.00000001.00000001.01000010 = 193.1.1.66/27
Host #3: 11000001.00000001.00000001.01000011 = 193.1.1.67/27
Host #4: 11000001.00000001.00000001.01000100 = 193.1.1.68/27
Host #5: 11000001.00000001.00000001.01000101 = 193.1.1.69/27
……………………………….……………………………………………….
Host #29: 11000001.00000001.00000001.01011101 = 193.1.1.93/27
Host #30: 11000001.00000001.00000001.010 11110 = 193.1.1.94/27
Example 1 – part 5

 Organisation has assigned the network number 193.1.1.0/24

and wants to have 6 subnets, where the largest one need to
support up to 25 hosts.

 Defining the Broadcast Address for Each Subnet

 The broadcast address for Subnet #2 is the all 1's host address, or
 11000001.00000001.00000001.01011111 = 193.1.1.95

 Note that the broadcast address for Subnet #2 is exactly one

less than the base address for Subnet #3 (193.1.1.96).
 This is always the case – the broadcast address for Subnet#n is one
less than the base address for Subnet #(n+1).
Example 2 – part 1

 Organisation has assigned the network number

140.25.0.16/16 and it needs to create a set of subnets
that supports up to 60 hosts on each subnet.

 Step one: number of bits required to define 60 hosts on

each subnet is 6 bits
 6 bits define 64-2=62 host address
 in order to take into account future growth we select 7 bits which
define 126 addresses per subnet.
 Determine the subnet mask/extended-prefix length:
Example 2 – part 2

 Organisation has assigned the network number

140.25.0.16/16 and it needs to create a set of subnets
that supports up to 60 hosts on each subnet.

 Step two: determine the subnet mask/extended-prefix

length
Example 2 – part 3

 Organisation has assigned the network number

140.25.0.16/16 and it needs to create a set of subnets that
supports up to 60 hosts on each subnet.
 Defining Each of the Subnet Numbers: the 9 bits allocated in
the subnet portion of the IP address allows 512 different
subnetworks.
Base Net: 10001100.00011001.00000000.00000000 = 140.25.0.0/16
Subnet 0: 10001100.00011001.00000000.00000000 = 140.25.0.0/25
Subnet 1: 10001100.00011001.00000000.10000000 = 140.25.0.128/25
Subnet 2: 10001100.00011001.00000001.00000000 = 140.25.1.0/25
Subnet 3: 10001100.00011001.00000001.10000000 = 140.25.1.128/25
Subnet 4: 10001100.00011001.00000010.00000000 = 140.25.2.0/25
…………………………………………………………………………………………..
Subnet 511: 10001100.00011001.11111111.10000000 = 140.25.255.128/25
Example 2 – part 4

 Organisation has assigned the network number 140.25.0.16/16 and it

needs to create a set of subnets that supports up to 60 hosts on each
subnet.
 Defining Hosts Addresses for Each Subnet: The 7 allocated bits will
give 126 different host ID.
 For subnet 3 for example we have:

Subnet #3: 10001100.00011001.00000001.10000000 = 140.25.1.128/25

Host 1: 10001100.00011001.00000001.10000001 = 140.25.1.129/25
Host 2: 10001100.00011001.00000001.10000010 = 140.25.1.130/25
Host 3: 10001100.00011001.00000001.10000011 = 140.25.1.131/25
…………………………………………………………………………………………
Host 126: 10001100.00011001.00000001.11111110 = 140.25.1.254/25
Example 2 – part 4

 Organisation has assigned the network number

140.25.0.16/16 and it needs to create a set of subnets that
supports up to 60 hosts on each subnet.

 Defining the Broadcast Address for Each Subnet:

 For subnet 3 again this address will be the following:
 10001100.00011001.00000001.11111111 = 140.25.1.255
Subnets and route aggregation

 KCL’s IP addresses could be divided by departments

KCL network: 137.73.0.0 /16

Informatics network: 137.73.8.0 /23
Informatics subnet ID is 0000100 (binary) or 4 (decimal)
A host in informatics: 137.73.9.232 /23
10001001.01001001.00001001.11101000

 The global/UK router needs to look at the first 16 bits to determine it

is a KCL address
 The KCL router needs to look at the first 23 bits to determine it is an
Informatics address
 The Informatics router needs to look at the last 9 bits (after the first 23 bits)
to determine which host the message is for
Problems with fixed length subnetting

 Different subnets can have different numbers of hosts,

e.g. Department of Informatics may have a lot more
computers than Department of English

 As the subnet ID gets longer, the number of bits left for

addressing hosts reduces
 For example:
 to distinguish 300 subnets, you need a subnet ID of 9 bits (as 28
= 256 and 29 = 512)
 If you have a /16 network and a subnet ID of 9 bits, this leaves 7
bits for host IDs, at most 126 hosts (27 – 2)
 What if a department needs 200 hosts?
Example subnetting problem

 A /24 network needs the following five subnets

 Subnets P, Q, R require 12 hosts each
 Subnet S requires 36 hosts
 Subnet T requires 90 hosts

 8 bits network prefix (32 – 24 = 8)

 254 available addresses in the network (28 – 2 = 254)
 The subnets need 172 addresses including network and broadcast
 If we use a 3 bit subnet ID
 We can accommodate all subnets, as 23 = 8 and we have 5 subnets
 But each has only 30 hosts, as 8 – 3 = 5 host ID bits, so 25 – 2 = 30 hosts
 If we use a 1 bit subnet ID
 Enough hosts per subnet, as 27 – 2 = 126 hosts
 But only 2 subnets, as 21 = 2
Variable length subnetting

 To solve this problem, we use variable length subnets

 Using variable subnet ID lengths, we iteratively divide up

the host ID space, first into large blocks, then into
smaller ones
 First, to accommodate the largest subnet, T, we use 1 bit to split
the T subnet from the rest
 We then accommodate the second largest subnet, S, by splitting
the remainder: S needs 6 bits for host IDs, so use 1 more bit to
split S from the rest
 Finally, we need 2 bits to split subnets P, Q and R
Variable length subnet masks

 As the number of bits used for the subnet prefix vary, so

the subnet masks vary for each variable length subnet
 Subnet T has addresses X.X.X.X /25
 24 bits for network prefix + 1 for subnet ID
 Subnet S has addresses X.X.X.X /26
 Subnets P, Q, R have addresses X.X.X.X /28
Variable length subnets example

 /24 network
 Subnets P, Q, R require 12 hosts each
 Subnet S requires 36 hosts
 Subnet T requires 90 hosts

 Network
 T : /25
 S : /26
 PQR: /28
Variable length subnets example

T: 10001001 01001001 00001001 0 . . . . . . .

137.73.9.0 /25
S: 10001001 01001001 00001001 10 . . . . . .
137.73.9.128 /26
P: 10001001 01001001 00001001 1100 . . . .
Q: 10001001 01001001 00001001 1101 . . . .
R: 10001001 01001001 00001001 1110 . . . .
137.73.9.192 /28
137.73.9.208 /28
137.73.9.224 /28
Outline

 Internet architecture

 Addressing hosts on the internet

 IPv4 addressing
 IPv6 addressing

 Routing Protocols
IPv6 header

 Although IPv6 Addresses are 4 times bigger (16 Bytes) than

IPv4 its Header is only twice the size of the IPv4 Header.
IPv6 addresses

 An IPv6 address uses 16 bytes / 128 bits

 All IPv6 local networks are /64, removing the need for variable
length subnetting at the local level
 Expressed in hexadecimal 16-bit words with :s
21DA:D3:0:2F3B:2AA:FF:FE28:9C5A
 Uses the same prefix length notation as IPv4:
21DA:D3:0:2F3B:2AA:FF:FE28:9C5A /64
 An IPv6 address identifies one or more interfaces rather than
hosts
 One network node may have many interfaces
 Each interface may be used to identify the node for a different
purpose or application
IPv6 address types

 IPv6 divides addresses into three types

 Unicast: Address for a single interface

 Anycast: Address for a group of hosts
 When a message is sent to an anycast address, it is delivered to
one of the hosts with that address
 Multicast: Address for a group of hosts
 When a message is sent to an multicast address, it is delivered
to all of the hosts with that address
Multicast and link-local

 Unlike IPv4, IPv6 does not restrict structure of addresses

(there are no address classes)
 However, there are two prefixes which have special
meanings:

 Multicast: 11111111
 Link-Local Unicast: 1111111010

 Multicast addresses are used for addressing a group of hosts

 There are no broadcast addresses for networks as in IPv4
 Link-local addresses are for use on a single link (i.e. in a
single local network), e.g. for automatic address configuration
Unspecified and loopback addresses

 Two special addresses may never be assigned to

interfaces

 Unspecified Address 0:0:0:0:0:0:0:0

 Indicates absence or no knowledge of an address

 Loopback Address 0:0:0:0:0:0:0:1

 Used by a host to send messages back to itself
Outline

 Internet architecture

 Addressing hosts on the internet

 IPv4 addressing
 IPv6 addressing

 Routing Protocols
 Shortest path routing
 Routing in the Internet
Routing and addresses

 Generally, one host wants to communicate data with

another host that is not directly connected

 To allow this to happen, data is routed:

 A path is found along a series of connected nodes
 Data is sent from one to the next along the path until reaching its
destination

 The best possible path for routing data to the destination

is found by the routing algorithm.
Routing

 Complex, crucial aspect of packet switched networks

 Characteristics required
 Correctness
 Simplicity
 Robustness
 Stability
 Fairness
 Optimality
 Efficiency
Performance Criteria

 Minimum hops

 Minimum cost
 Each link has a cost that reflects
 The length of the link
 Delay on the link
 Congestion
 Cost
 Cost may change with time
 Examples:
 Bellman-Ford
 Dijkstra’s algorithm
 Many others
Decision Time and Place

 Routes must be chosen for various source destination

pairs (S/D pairs) or for various sessions
 Decision Time
 Datagram routing: route chosen on a packet by packet basis
 Using datagram routing is an easy way to split paths
 Virtual circuit routing: route chosen a session by session basis
 Static routing: route chosen in a prearranged way based on S/D
pairs
 Decision place
 Centralised
 Distributed
Routing Strategy - Fixed

 Fixed Routing
 Single permanent route for each source, destination pair
 Determine routes using a least cost algorithm
 Dijkstra Algorithm
 Bellman-Ford Algorithm
 Route fixed, at least until a change in network topology
Routing Strategy - Flooding

 Flooding
 No network info required
 Packet sent by node to every neighbour
 Incoming packets retransmitted on every link except incoming link
 Eventually a number of copies will arrive at destination
 Each packet is uniquely numbered so duplicates can be discarded
 Can include a hop count in packets

 Properties:
 All possible routes are tried
 Very robust
 At least one packet will have taken minimum hop count route
 All nodes are visited
 Useful to distribute information (e.g. routing)
Routing Strategy - Random

 Random
 Node selects one outgoing path for the retransmission of
incoming packets
 Selection can be random or round robin
 Outgoing path can be selected based on probability calculation
 No network info needed
 Route is typically not least cost nor minimum hop
Routing Strategy - Adaptive

 Adaptive
 Used by almost all packet switching networks
 Routing decisions change as conditions on the network change
 Failure
 Congestion
 Requires info about network
 Decisions are more complex
 Trade-off between quality of network info and overhead
 Reacting too quickly can cause oscillation
 Reacting too slowly, information may not be relevant
Outline – Next week

 Internet architecture

 Addressing hosts on the internet

 IPv4 addressing
 IPv6 addressing

 Routing Protocols
 Shortest path routing

 The Internet Protocol