Network Layer

provides logical communication between hosts (not processes running on host) and doesn’t know the actual process being sent as that is responsible by transport layer

• data plane

• control plane

contains the forwarding function:

multiple per-router functions in network layer that determine how a datagram arriving at one router’s input link is forwarded to the router’s output link

functions include -

• forwarding packets

• blocking packets from exiting a router (eg. malicious host or to a forbidden destination host)

• duplicating packets and sending them over multiple outgoing links

contains the routing protocol:

protocols that include network-wide logic that controls how data is routed amongst routes along an end-to-end path (from source to destination host) using routing algorithms

• traditional - data plane forwarding function and control plane routing protocol are implemented together within a router

• software-defined networking (SDN) - separates data plane and control plane. data plane in router and control plane in a separate service, eg. remote controller

in hosts and in routers

sending side ⇒ encapsulates segments from transport layer into datagrams (with network header) and sends to router

receiving side ⇒ receive datagram from nearest router, extract transport layer segment (by unwrapping the network datagram and removing network header) and sends segment to transport layer

forwarding

routing

forwarding is a router-local action which transfer packet from the router’s input link interface to its output link interface - takes nanoseconds and is implemented in hardware

• actually moving the packet

• eg. packet arriving at a nearby router needs to be forwarded to the next router on the path

routing is a network-wide process to determine end-to-end path that a packet takes - takes seconds and is implemented in software

• network layer defines the route for the packet to take to arrive at the destination host

• uses routing algorithms to define the best route (shortest, least cost etc.) and sends the forwarding information to the router

using forwarding tables - every router has a forwarding table

the routing algorithm determines the content of the router’s forwarding table

1. the router examines the values of certain fields in the arriving packet header

2. uses the header values to get an index to use in the forwarding table

3. value stored at forwarding table[index] indicates the outgoing link interface that the router needs to forward the packet to

• the routing algorithm is in every router and runs to find the best route

• router contains forwarding and routing functions

• routing algorithm functions in one router communicate with the routing algorithms in another to compute the values for forwarding table

• the different algorithms in each router communicate by exchanging routing messages with routing information that follow a routing protocol

• routing algorithm implemented centrally, remotely in a remote controller (eg. implemented as a data centre with high reliability + redundancy, managed by ISP or third party)

• the remote controller distributes the forwarding table to every router

• routing device (router) only does forwarding, remote controller computes and distributes the table

• routers and remote controller communicate by exchanging messages containing the forwarding tables and other routing info

contains 4 components

1. input ports

2. switching fabric

3. output ports

4. routing processor

to receive packets

• performs physical layer function - terminates incoming physical link at a router

• performs link layer function - gets packet from link layer and transports it to network layer, interoperate with link layer at other side of incoming link

• performs the lookup function on the forwarding table to determine the router output port that the packet needs to be forwarded to

• control packets are forwarded from the input port to the routing processor

• the number of ports supported by the router can vary

where switching occurs

• connects the router’s input ports to its output ports

• completely contained in router

to send packets

• stores packets received from switching fabric

• transits packets on the outgoing link by performing link layer + physical layer functions

performs control plane functions (routing) and network management functions

• traditional routers - executes routing protocols, maintains routing tables and attached link state information, computes the forwarding table for that router

• SDN routers - communicates with the remote controller to receive forwarding table entries computed by the remote controller and installs these entries in the router’s input ports.

used to find paths that packet needs to travel

path - sequence of routers the packet will traverse through from source to destination

good path = least cost, fastest, least congested

• shortest

• fastest

but the shortest and fastest path might not be the same

• centralised

• decentralised

• static

• dynamic

• load sensitive

• load insensitive

all routers have the complete network topology (used link state algorithms)

routers only knows the physical connection to the neighbouring routers (uses distance vector algorithms)

routes change slowly over time

can change based on what information we know (eg. traffic updates, topology changes)

link costs vary dynamically to reflect congestion level in underlying link

link cost does not reflect the past or current level of congestion

IPv4

IPv6

network layer

link layer

physical layer

• header length

• length

• header checksum

• options

• data/ payload

• source IP

• destination IP

• version number

• type of service

• identifier

• flags

• fragmentation offset

• time to live

• upper layer

datagram = segment + network header

5-2-7

for IPv4 or IPv6

4 bits

same as SEGMENT: header can vary in length because of optional fields, typical header length is 20 bytes

4 bits

distinguish between different types of IP datagrams and request services (real time, non-real time/ priority, high throughput)

[8 bits]

header length + data length

[16 bits]

contain information used for IP fragmentation and reassembly

ensures datagram doesn’t live forever as there could could be errors, deadlock that causes processing time. each time datagram is processed by router TTL--, when TTL = 0, router drops datagram

information about what the upper layer is: transport layer, TCP and UDP

electronic components can cause bit corruption and routers discard datagrams with detected errors. checksum must be recomputed and stored at each router since TTL + other fields can change

when a source creates a datagram, it inserts its IP address into the source IP address field and inserts the address of the ultimate destination into the destination IP address field

optional fields that extends the header length

transport layer segment (either TCP segment or UDP segment) or other type of data eg. ICMP message

• 20 bytes of datagram header

• payload { 20 byes of transport layer header, application layer message }

• in total- 40 bytes of header and an application layer message of variable length

need to know length of datagram because it needs to be transmitted through a link and the link layer frame can have size restraints

1. the IP datagram is encapsulated within the link layer frame for transport between one router to another

2. link layer protocols can differ and use different protocols so all can’t carry the same size of network packets

3. each link layer frame has a variable: MTU maximum transmission unit which restricts IP datagram size

4. if datagram size > MTU ⇒ fragmentation occurs

breaking down payload into smaller IP datagrams called where each datagram has its own header, broken down based on the MTU

1. each datagram has an identification number assigned by the host

2. each fragment of a datagram has the same source address, destination address and identification number as the original datagram

3. when the destination receives the fragments, it can look at the identification number to show which fragments are part of the same datagram

4. IP is unreliable and thus not guaranteed that the fragments will be sent in the same order or correctly

flag

offset

each fragment will have a flag bit - destination hosts needs to ensure that all fragments that make up a full datagram have been received

• all fragments have flag bit = 1

• the last fragment has flag bit = 0

each fragment has an offset variable - offset specifies where there fragment fits in the original datagram so that the destination host can reassemble in the correct order + determine any missing fragments

• starts at 0 and is split into 8 byte blocks

• host has normally 1 link to the network

• router has 2 or more links to the network

every host/ router can send/ receive IP datagrams so each interface needs to have an IP address

interface = the boundary between both the host/router and any one of their links

IP address related to the host/ router interface rather than the actual host/ router

32 bits long

dotted decimal notation - each byte of the address is written in decimal form separated with a dot

cannot be chosen randomly

• subnet.subnet.subnet.host/subnetmask

• part of the IP address shows what subnet it is connected to

small section of a network, formed by a network interconnecting many host interfaces and one router interface

IP addressing assigns an address to a subnet

/XX indicated that the leftmost XX bits of the total 32 bits define the subnet address

eg. /24 is the submnet mask - 24 left-most bits belong to the network

there are different network classes based on what subnet mask are used

Class A subnet mask /8 - Class B subnet mask /16 and Class C subnet mask /24

/8 subnet vs. /24 subnet - 8 bits used, so the rest of the bits can be used to specify the hosts - so more IP addresses available, a bigger network that a /24 subnet

subnet.X.X.0 - the .0 means that you want to access all the interfaces on that subnet (network)

Classless Interdomain Routing CIDR [RFC 4632]

ip addressing format - 32 bits etc.

using /xx for the subnet mask

range of IP addresses with common prefixes

a new type of IP addressing with extended addressing capabilities needed as 32 bit IPv4 address space is being used up

expanded addressing capabilities

• 128 bits long - ensures we won’t run out of ip addresses,

• unicast, multicast and new types of addresses (anycast address) that allows datagram to be delivered to an one of a group of hosts

fixed 40 byte header

• the header has no optional fields that are in IPv4

• faster processing of datagram by router