Chapter 4 Full

11/28/2016
Digital Routing
ROUTING IN IP NETWORKS
11/28/2016
DR. PHUONG NGA TRAN - EEIT
Content
Introduction
Single path routing
Broadcast
11/28/2016
11/28/2016
Introduction
Routing determines the communication paths to be used for
the communication between any two parties.
We consider traffic flow 1 4 and 3 6.
Direct link
No routing problem as long as traffic intensities (load) are

below the link capacity (5)
1
11/28/2016
6
3
Introduction
Routing classification:
Single path routing
Multipath (bifurcated routing)
Broadcast
1
5
5
5
5
4
11/28/2016
5
6
4
11/28/2016
Introduction
Routing decisions are implemented as routing tables in
network nodes.
A distinction has to be made whether or not we are
considering:
a connectionless or
a connection-oriented service.
For datagram operation the packet contains the full
destination address in the packet header.
The routing table:

contains all possible destination addresses as entries;
the content is the port to which the packet should be output.
11/28/2016
Routing table
2
Destination node
Next node
Routing table for node 4

11/28/2016
11/28/2016
Routing table
The entries have been generated upon evaluation of a routing
algorithm.
Idea: Try to optimize some cost function for effective use of
resources in the network.
Performance criterion
Euclidian distance
number of hops
throughput
delay
costs (e.g. for the utilization of transmission links)
11/28/2016
Routing Algorithm
Major concerns when developing such an algorithm:
correctness
complexity
stability
fairness
Information used in the algorithm
none
only local information
from adjacent nodes
from a subset of nodes
from all nodes
11/28/2016
11/28/2016
Routing Algorithm
Routing strategy
Decision place
static
each node
adaptive, updates:
central node
continuous
source node
periodic
subset of nodes
upon major load change

upon topology change
Decision time
packet (datagram)
session (virtual circuit)
11/28/2016
10
Algorithms to determine shortest paths

These algorithms in general try to minimize a cost function.
"Costs" can be anything which is sensible to minimize as
seen by the operator of the network, e.g.:
Euclidian distance
number of hops
tariffs for links, etc.
link utilization
or any other measure derived from the above.
The costs are then condensed in a cost matrix obeying the

following rules:
cii = 0
cij = ; if i and j are not adjacent nodes.
11/28/2016
10
11/28/2016
11

5
3
3
Network topology
Note! The cost matrix is in general symmetric with
respect to its main diagonal.
Cost matrix
11/28/2016
11
12

An optimal route (shortest path) from a node 1 to a node k is a
sequence of links (pq): 1a,..., ji ,..., mk
such that
costs Min
links
All algorithms make use of Bellmann's optimality principle:
Assume that the optimal path S1k from node 1 to node k has been
found.
Let i be a node on this path; then S1i and Sik are also optimal!
11/28/2016
12
11/28/2016
13

We conclude:
Let
be the minimal distance from node 1 to node .
Let
be the minimal distance to node with j being a direct
neighbor of
From Bellmann's principle we then derive the following equations:
D1 0
D i min [ D j c ji ], i 1
j N ( i )
( ) is the set of neighbours of .

The equation lends itself to an iteration and different algorithms
differ in the way of how the iteration is organized.
11/28/2016
13
14
Dijkstra's Algorithm
Assume that we want to find shortest path from node 1 to all
other nodes
(p,d)= (predecessor,
distance to 1)
M = {}
Step 1
Initialization
5
(-1,)
(-1,)
3
3
(-1,)
(-1,0)
1
11/28/2016
(-1,)
(-1,)
114
11/28/2016
15
Min Dj = D1=0
M={1}
Dj=min (,0+c1j)= c1j
Step 2
After first iteration
5
(1,5)
(1,2)
3
(-1,)
1
(-1,0)
1
4
(1,1)
11/28/2016
(-1,)
15
16
Consider node j, which has the lowest distance to 1 so far
Step 3
After second iteration
5
(4,4)
(1,2)
3
Min Dj = D4=1
M={1,4}
D2=min (2,1+2) = 2
D3=min (5,1+3) = 4
D5=min (,1+1)= 2
(-1,)
1
(-1,0)
1
4
(1,1)
11/28/2016
5
(4,2)
16
11/28/2016
17
Min Dj = D2=2
M={1,2,4}
D3=min (4,2+3) = 4
Step 4
After third iteration
5
(4,4)
(1,2)
3
(-1,)
1
(-1,0)
1
4
(1,1)
11/28/2016
(4,2)
17
18
Step 5
After fourth iteration
5
(5,3)
(1,2)
Min Dj = D5=2
M={1,2,4,5}
D3=min (4,2+1) = 3
D6=min (,2+2) = 4
(5,4)
1
(-1,0)
1
4
(1,1)
11/28/2016
(4,2)
18
11/28/2016
19
Min Dj = D63=4
=3
M={1,2,3,4,5}
M={1,2,3,4,5,6}
D6=min (4,3+5) = 4
Step 7
6
After sixth
fifth iteration
iteration
5
(5,3)
(1,2)
3
(5,4)
1
(-1,0)
1
4
(1,1)
11/28/2016
(4,2)
19
20
Summary of Dijkstra's Algorithm

The Dijkstra algorithm will determine the shortest path from a source
to (N-1) destinations and iterates over the path length.
Starting from the source node (1) the nearest neighbour (minimal c1i)
is determined.
Then from the nearest neighbours of these two nodes that one with
the shortest distance (SD) to the source is selected.
Starting now with three nodes, again among their neighbours that one
with the SD to the source is looked for, etc.
The elected nodes with SD to the source are collected in a set M.
The algorithm terminates when the set V of all nodes is identical to M.
We update labels (p;d) for each node indicating the actual values of
predecessor node and
actual distance to node 1.
11/28/2016
10
11/28/2016
21
Pseudo code of Dijkstra's Algorithm
N: number of nodes
V: set of all nodes
M: set of considered nodes
11/28/2016
22
Dijkstra's algorithm or variants of it are implemented in the
INTERNET.
Its complexity is 0(N2):
We have N-1 iterations each containing a number of operations
which is proportional to N.
If all cij are known to all nodes, each node can execute the
algorithm independently.
To proceed like this, changes in the cost matrix need to be
communicated in the whole network.
11/28/2016
11
11/28/2016
Exercise
A network topology and the link costs are given as in the figure below.
Node A is using the Dijkstra algorithm to compute its routing table. The
following table illustrates the iterations of the algorithm. Please complete
it given that (p; d) = (predecessor; distance to A).
Iteration
B
C
D
E
Initialization (-1; ) (-1; ) (-1; ) (-1; )
1
2
3
4
M
{}
3
E
3
1
M = set of considered nodes
11/28/2016
23
Bellman Ford Algorithm

More general than Dijkstras algorithm:
Link cost can be negative
Detects the existence of negative-cost cycle(s) reachable
from a single source.
If one negative cost cycle reachable from the source exists, there
is no cheapest path.
Applied for directed graphs
11/28/2016
24
12
11/28/2016
Bellman-Ford Algorithm
INITIALIZATION
(1) = 1
(1) = 0
for
= 2
RELAX EDGES REPEATEDLY

=1
for
1 do
for each edge ( , ) do
do
if
<
=
= 1
() =
CHECK FOR NEGATIVE-COST CYCLE
for each edge ( , ) do
N: number of nodes
d(i): distance from I to the source
p(i): predecessor node of i on the
path from the source to i
if
<
error graph contains negative cycle
11/28/2016
25
5
2
6
1
3
-2
-3
-4
2
Sequence of considered links: (1-2); (2-3); (4-3); (5-4); (5-3); (2-5); (4-1); (1-5); (2-4); (3-2)
11/28/2016
26
13
11/28/2016
5
2
6
1
(Source)
11
-3
-2
-4
2
7
7
11/28/2016
27
5
2
6
1
(Source)
-3
-2
7
7
-4
11/28/2016
28
14
11/28/2016
5
2
6
1
(Source)
-3
-2
7
7
-4
-2
11/28/2016
29
5
2
6
1
(Source)
-3
-2
7
7
-4
-2
11/28/2016
30
15
11/28/2016
Exercise
Use Bellman Ford algorithm to compute the shortest path from node 1 to all other nodes
3
2
1 0
Source
11/28/2016
31
Proof of Bellman Ford Algorithm

Lemma 1: Upper Bound Property
( )
We always have ( ) ( , ) and if we ever find

=
never changes ( , is the shortest distance from to ).
, ( )
Proof:
At iteration 0,
= 0,
Assume that after k iterations,
( ) and
( )
At iteration k+1
- By triangle inequality,
- If ( ) = ( ) +
( )
( )+
If we ever achieve ( ) = ( , ), we cannot decrease ( ) since we have

just shown that ( ) ( , ). Moreover, we cannot increase ( ) because
we only change d[v] at an iteration if ( ) decreases. Therefore, once we have
( ) = ( , ), it cannot change.
11/28/2016
32
16
11/28/2016

Lemma 2: Path relaxation
Assume we have a graph G with no negative cycles. Let = ( , , , . . ) be a
shortest path from to . If edges are relaxed in the order
( , ), ( , ), , (
, ),
will be the shortest distance from to
(
=
0,
) after all of these relaxations and at all times afterwards. Note
that this property holds regardless of what other relaxation calls are made before,
during, or after these relaxations.
Proof:
Base case: at iteration 0 (initialization) ( ) = 0 = ( , )
Assume that we have relaxed, in order, edges ( ,
= ( ,
).
We have
Before edge (
=
,
after relaxing edge

11/28/2016
) is relaxed,
,
), ( ,
), , (
), and
because p is the shortest path.
(upper bound property)
,
,
33

Lemma 3: Bellman-Ford Correctness
Assume we have a graph with no negative cycles reachable by . Then after
running Bellman-Ford, we have ( ) = ( , ) for all reachable from .
Proof:
If is reachable from then there must exist an acyclic shortest path =
( , , , . . ) where
= . Now, since is acyclic, can contain no more
than vertices and therefore no more than | | 1 edges.
At each iteration of Bellman-Ford, we relax every edge. Therefore, we relax
( , ) on the first iteration, ( , ) on the second iteration, etc (we also relax
( , )on the first iteration, of course, but we cannot guarantee that relaxation
comes after the relaxation of( , ). By the time we have reached the | |
1 iteration, we must have relaxed every edge in in order. Therefore, by the
path relaxation property, we have ( ) = ( , ) when we have done | |
1 iterations
11/28/2016
34
17
11/28/2016
Summary of Bellman Ford Algorithm

Like Dijkstra algorithm, Bellman Ford is based on the principle
of relaxation, in which an approximation to the correct distance is
gradually replaced by more accurate values until eventually
reaching the optimum solution.
Bellman Ford algorithm simply relaxes all the edges and does
this 1 times ( is the number of nodes).
In each repetition, the number of nodes with correctly calculated
distances grows, from which it follows that eventually all nodes will
have their correct distances.
BellmanFord algorithm can be applied to a wider class of inputs
than Dijkstra, i.e. negative link cost.
The complexity of Bellman Ford algorithm is ( )
11/28/2016
35
36
Distributed Routing
In a network with distributed routing each node follows a
routing algorithm (e.g. Dijkstra's) and creates his local routing
table.
This information is periodically sent to all neighbours.
After some time each node knows the shortest path to all
destinations.
The periodic updates guarantee the adaptability to failing
links or nodes.
11/28/2016
18
11/28/2016
Distributed routing
Distance vector routing protocol
Link state routing protocol
11/28/2016
37
Distance vector routing

Each node constructs a one-dimensional array containing the
distances or costs to all other nodes (as it relates to its
knowledge) and distributes it to its immediate neighbors.
Key thing -- each node knows the cost of links to its
neighbors.
If no link exists between two nodes, the cost of a direct link
between the nodes is infinity.
11/28/2016
38
19
11/28/2016
39

In distance-vector routing the information sent to the
neighbours is a set of pairs (destination, distance) of all other
nodes.
The receiving node checks:
whether for a given destination the advertised distance agrees
with its own findings or
whether a routing via the sender of the information is
advantageous.
Distance vector routing uses Bellman
11/28/2016
Ford algorithm
Example
A
B
A
C
D
E
F
1
0
Internal Information at each node
11/28/2016
40
20
11/28/2016
Example
Cost
Next
Hop
B
A
C
D
E
F
With the initial information,

routing table at A is as above
11/28/2016
41
Example
Each node sends a message to neighbors
with a list of distances.
Once a node receives a message from its
neighbours, it updates the distance as
follow
= min{
B
A
C
D
E
F
11/28/2016
Cost
Next
Hop
G
42
21
11/28/2016
Convergence
In the absence of topological changes -- few exchanges
between neighbors before complete routing table is formed.
This table is consistent.
Convergence is achieved.
Notice -- no centralized authority
11/28/2016
43
Routing updates
When are routing updates sent ?
Periodic updates
Even if nothing has changed, send periodically. Main
reason is to let other nodes know that the sender is alive.
Refresh information that might be needed if some of the
routes were to become unavailable.
Triggered updates
When a node receives an update from one of its
neighbors which may lead to a change in its routing tables
(could be due to change in link cost).
Note: typically order of periodicity is seconds to several

minutes.
11/28/2016
44
22
11/28/2016
Link/Node Failures
Nodes that first notice send new lists of distances to neighbors.
How do they detect failures ?
Route updates dont arrive
Probing with test packets.
11/28/2016
45
Example revisited
Let link from F to G fail.
F sets new distance to G to ; sends update to A.
A was ini ally rou ng to G via F. So it now sets link cost to G to .
Next update from C; A learns that C has 2 hop path to G.
A now can reach G in 3 hops via C.
A sends an update to F. Thus, F now, can reach G via A in 4 hops.
B
A
C
D
E
F
11/28/2016
G
46
23
11/28/2016
Count to infinity problem

A discovers that link to E is lost.
The process continues

and thus, the system
does not stabilize.
If before As message (saying that link cost to

E is ) is received, if B or C adver se that
they can reach E in two hops, then A can be
confused.
This is the count to

infinity problem.
Another possibility, B gets As update

followed by Cs update which says that E is
reachable in 2 hops.
So B tells A this, and A thinks it can now
reach E via B in 3 hops.
This information reaches C who now thinks
that it can reach E in 4 hops via A.
11/28/2016
B
A
C
D
E
F
47
Solution for Count to Infinity Problem

Split Horizon
With Poison reverse
With Split Horizon, when a node sends a routing table

update to its neighbors, it does not send those routes it
learned from a particular neighbor, back to that neighbor.
For example, node A had a route to D via C with the distance of
2. When it sends a route update to C, it does not include this.
With split horizon with poison reverse, this update is

reported but the link weight is set to .
For example A sends (D, ) to C.
11/28/2016
48
24
11/28/2016
Link State Routing

Initial state : similar to distance vector i.e., state of link to
neighbors known (up/down).
Goal: To find the path of least cost to destination.
Basic Idea -- Every node knows how to reach its neighbors.
If this info is dissemination to every node, every node
ultimately has the information to build the complete map
of the network.
11/28/2016
49
Link State Routing - Mechanism

Two mechanisms:
Reliable dissemination of link state information -- process
is called reliable flooding.
calculation of routes using the collected information -- the
computation is based on Dijkstras algorithm.
11/28/2016
50
25
11/28/2016
Reliable flooding
Process of making sure that all the nodes
participating in the link state routing protocol get a
copy of the link-state information from all other
nodes.
Each node sends out link-state information on its
directly connected links.
Each node that receives this, forwards it.
11/28/2016
51
Link State Information

Each node creates a link-state packet (LSP) that contains:
ID of the node that created LSP
a list of directly connected nodes and the cost to each node.
sequence number
TTL
11/28/2016
for reliability
52
26
11/28/2016
Example
X receives LSP from some node Y.
X checks to see if it already has
an update from Y. If it does, it
compares the sequence number
in the new LSP to the one stored.
If New seq no < Old sequence
number, then, discard LSP.
Else -- store LSP and send the LSP
to all neighbors except the one
that sent the LSP.
(a)
(b)
(c)
(d)
If no update from Y, keep it.
11/28/2016
53
Dissemination of LSPs
LSPs are sent periodically (upon the expiry of a timer) or may be
triggered due to a change in topology
The only topology change that triggers the creation of a new LSP is a
change to one of the directly connected links.
Failures detected by link layer protocol by using what are known as HELLO
packets -- probes to determine if neighbor is alive.
To minimize overhead, LPSs are not created unless needed -periodicity is of the order of hours.
Sequence numbers help in identifying new info and TTL helps in
ensuring that packets dont stay in the network indefinitely.
11/28/2016
54
27
11/28/2016
55
Stability and convergence problems

If the distances (costs) are traffic intensity dependent,
shortest path routing will have stability problems.
However, it makes sense to route traffic to links which are
less loaded than others!
11/28/2016
56

Example:
D i
i
Destination
1
Step 1
Initial routing
3+
2+
2
2
1+
11/28/2016
28
11/28/2016
57

Example:
D i
i
Destination
1
Step 2
First rerouting
8
1
4+
3+
2+
3
1+
11/28/2016
58

Example:
D i
i
Destination
1
Step 3
Second rerouting
6+
5+
1
2
4+
11/28/2016
3+
29
11/28/2016
59

Example:
D i
i
Destination
1
Step 4
Third rerouting
8
1
6+
5+
4+
3
1
11/28/2016
3+
Oscillatory Routing!
60

To damp oscillations one strategy could be to stagger the
rerouting times in such a way that only a subset of nodes is
allowed to change the routing
11/28/2016
30
11/28/2016
Comparing Distance Vector and Link State

In distance vector depends on relative path costs -- count to
infinity problem.
However, link state suffers from oscillations.
Link state is stable since computation done at each node -distance vector could lead to problems if there are malfunctioning
routers.
11/28/2016
61
Broadcast Routing
Broadcast applications exist:
in the management of networks, e.g. updates of databases and
routing tables;
at application level e.g. mail applications or radio programs.
A broadcast method that does not need any new feature is to send a
distinct packet to each user. Seen from the resource consumption this
is the least desirable method.
A more suitable method might be flooding where a packet is output
to every port besides the one it entered the node.
Obviously flooding needs a termination criterion, e.g. a hop count.
Flooding also suffers from an abundant generation of packets.
11/28/2016
62
31
11/28/2016
Broadcast Routing
Flooding example
5
7
6
4
2
11
10
13
12
11/28/2016
63
Broadcast Routing
A more appropriate method is to use a spanning tree.
A spanning tree is a subset of a network that includes all nodes but
contains no loops
B
E
H
D
F
G
I
L
O
K
M
Each node must know some spanning tree.

With link state routing easily available!
With distance vector routing alone - not available!
11/28/2016
64
32
11/28/2016
Broadcast Routing
Reverse path forwarding: When a broadcast packet arrives at a router,
the router checks if the packet arrived on the port that is normally
used for sending the packet to the source of the broadcast (in our
example: I).
If so, it is forwarded to all other ports.
Otherwise the packet is discarded.
B
L
O
K
M
Assume hop count metric
11/28/2016
65
Routing in Global Internet

A routing protocol allows one router to exchange routing
information with another
However, such a scheme cannot scale to the entire Internet
if a router in the Internet attempted to exchange routing

information with all other routers, the resulting traffic would
overwhelm the Internet
To limit routing traffic
The Internet uses a routing hierarchy

Routers and networks in the Internet are divided into groups
All routers within a group exchange routing information
Then, at least one router (possibly more) in each group
summarizes the information before passing it to other groups
11/28/2016
66
33
11/28/2016

How large is a group?
What protocol do routers use within a group?

What protocol do routers use between groups?
The designers of the Internet routing system did not dictate an exact
size nor did they specify an exact data representation or protocol
Instead, the designers purposefully kept the architecture flexible
enough to handle a wide variety of organizations
For example, to accommodate organizations of various size
the designers avoided specifying a minimum or maximum size for a
group
To accommodate arbitrary routing protocols
the designers decided to permit each organization to choose a routing
protocol independently
11/28/2016
67
Autonomous System 1
Autonomous System 3
Autonomous System 3
11/28/2016
68
34
11/28/2016
Autonomous System
Term Autonomous System (AS) to specify groups of routers
One can think of an AS as a contiguous set of networks and routers
all under control of one administrative authority
There is no exact meaning for administrative authority
The term is sufficiently flexible to accommodate many possibilities
For example, an AS can correspond to an ISP, an entire corporation, or
a university
Alternatively, a large organization with multiple sites may choose to
define one AS for each site
In particular, each ISP is usually a single AS, but it is possible for a large
ISP to divide itself into multiple ASs
The choice of AS size can be made for

economic, technical, or administrative reasons
11/28/2016
69
Autonomous System
For example
consider a multi-national corporation
It may be less expensive for the corporation to divide into multiple
ASs, each of which has a connection to an ISP in a given country,
than to act as a single AS with one connection to the rest of the
Internet
Another reason for a specific size arises from the routing protocol
to be used
a protocol may generate excessive routing traffic when used on
many routers
(i.e., routing traffic may grow as the square of the number of
routers)
11/28/2016
70
35
11/28/2016
Types of Internet Routing Protocols
Some typical routing protocols. There are in practice much more.
11/28/2016
71
Interior Gateway Protocols (IGPs)

Routers within an Autonomous System (AS) use an IGP
exchange routing information
Several IGPs are available
each AS is free to choose its own IGP
Usually, an IGP is easy to install and operate
IGP may limit the size or routing complexity of an AS
11/28/2016
72
36
11/28/2016
Exterior Gateway Protocols (EGPs)

A router in one AS uses an EGP to exchange routing
information with a router in another AS
EGPs are more complex to install and operate than IGPs
but EGPs offer more flexibility and lower overhead (i.e.,
less traffic)
To save traffic
an EGP summarizes routing information from an AS before
passing it to another AS
An EGP implements policy constraints
that allow a system manager to determine exactly what
information is released outside the organization
11/28/2016
73
Illustration of IGPs and EGPs

IGP 3
IGP 1
EGP
Autonomous System 1
Autonomous System 3
IGP 2
Autonomous System 3
two-level routing hierarchy used in the Internet

11/28/2016
74
37
11/28/2016
Illustration of IGPs and EGPs

AS1 has chosen IGP1 to use internally, and AS2 has chosen IGP2
All routers in AS1 communicate using IGP1
All routers in AS2 communicate using IGP2
Border routers use an EGP to communicate between the two
ASs
That is, a border router must summarize information from its
AS and sends the summary to a border router of another AS.
In addition, the received summary is propagated to all
routers in an AS using its own IGP
11/28/2016
75
IGPs vs. EGPs

IGPs and EGPs differ in an important way with respect to routing
metrics:
IGPs use routing metrics, but EGPs do not
each AS chooses a routing metric and arranges internal routing software
to send the metric with each route so receiver can use the metric to
choose optimal paths
Each AS is free to choose a routing metric

Outside an AS, an EGP does not attempt to choose an optimal path
Instead, the EGP merely finds a path
An EGP cannot make meaningful comparisons
Suppose one AS reports the number of hops along a path to destination D

and another AS reports the throughput along a different path to D
An EGP that receives the two reports cannot choose which of the two
paths has least cost because there is no way to convert from hops to
throughput
Thus, an EGP can only report the existence of a path and not its cost
11/28/2016
76
38
11/28/2016
Border Gateway Protocol (BGP)

BGP is the most widely used EGP in the Internet
Goal of BGP is modest:
any loop free path is to be found between the source and destination
Routing Among AS
Because it is intended for use as an EGP
BGP provides routing information at the AS level
No way for BGP to provide details about the routers within each AS on
the path
Provision for Policies
BGP allows the sender and receiver to enforce policies
a manager can configure BGP to restrict which routes BGP advertises
to outsiders
11/28/2016
77

Reliable Transport
BGP uses TCP for all communications
a BGP program on a router in one AS forms a TCP connection to a
BGP program on a router in another AS and then sends data across
the connection
BGP provides the glue that holds Internet routing together

at the center of the Internet
Tier-1 ISPs use BGP to exchange routing information and learn
about each other's customers
11/28/2016
78
39
11/28/2016

Facilities for Transit Routing
BGP classifies each AS
as a transit system if it agrees to pass traffic through to another AS or
as a stub system if it does not
Traffic passing through on its way to another AS is classified as

transit traffic
Classification allows BGP to distinguish between ISPs and other
AS
BGP allows a corporation to classify itself as a stub
even if it is multi-homed (i.e., a corporation with multiple external
connections can refuse to accept transit traffic)
11/28/2016
79
BGP in details
BGP supports flexibility -- paths could be chosen by a
provider based on a policy.
To configure BGP, each AS admin picks at least one node to
be the BGP speaker -- a spokesperson node for the entire
AS.
The BGP speaker establishes a BGP session with other BGP
speakers in other ASes.
Source advertises complete paths (unlike distance vector or

link state routing) -- thus loops are prevented.
11/28/2016
80
40
11/28/2016
Example
AS 2 says 128.96, 192.4.15, 192.4.32, 192.4.3 can be reached
via AS 2.
AS 1 advertises that these networks can be reached via <AS1,
AS2> --note full path description.
Loops are avoided.
Regional provider A
(AS 2)
Backbone network
(AS 1)
Regional provider B
(AS 3)
11/28/2016
Customer P
(AS 4)
128.96
192.4.153
Customer Q
(AS 5)
192.4.32
192.4.3
Customer R
(AS 6)
192.12.69
Customer S
(AS 7)
192.4.54
192.4.23
81
BGP path attributes

In many cases, there will be multiple routes to the same
destination. BGP therefore uses path attributes to decide how
to route traffic to specific networks.
The easiest of these to understand is Shortest AS_Path. What
this means is the path which traverses the least number of AS
"wins."
Another important attribute is Multi_Exit_Disc (Multi-exit
discriminator, or MED). This makes it possible to tell a remote
AS that if there are multiple exit points on to your network, a
specific exit point is preferred.
11/28/2016
82
41
11/28/2016
BGP Messages
BGP has four types of messages
OPEN: Establish a connection with a BGP peer
Note: BGP connection is TCP based ! (Port no. 179).
UPDATE -- advertise or withdraw routes to a destination

Note --BGP speaker needs to be able to cancel previously
advertised paths if nodes or links fail. This form of negative

advertisements are said to advertise withdrawn routes.
KEEPALIVE: Inform a peer that the sender is still alive but

has no information to send.
NOTIFICATION: Notify that errors are detected.
11/28/2016
83
Routing with BGP

For stub AS -- border router injects a default route into the
intra-domain routing protocol.
If there are more than one border router, each injects
specific routes that they have learned from outside the AS.
IBGP or Interior BGP is used to distribute the information to
all other routers in the domain (and the speaker).
11/28/2016
84
42
11/28/2016
BGP issues
Routing table growth.
If a network operator decides to deaggregate a network that used to be a
single into multiple networks, they could potentially start advertising hundreds
of new routes. Every router on the Internet will get every new route when this
happens.
Someone will "advertise the Internet."

If some large ISP's customer suddenly decides to advertise everything, and the
ISP accepts the routes, all of the Internet's traffic will be sent to the small
customer's AS.
A solution is route filtering. It's quite simple to set up filters so that your
routers won't accept routes from customers that you aren't expecting, but
many large ISPs will still accept the equivalent of "default" from peers that
have no likelihood of being able to provide transit.
11/28/2016
85
BGP issues
Flapping.
BGP has a mechanism to "hold down" routes that appear to be flaky.
Routes that flap, or come and go, usually aren't reliable enough to send
traffic to. If routes flap frequently, the load on all Internet routes will
increase due to the processing of updates every time someone
disappears and reappears.
Dampening will prevent BGP peers from listening to all routing updates
from flapping peers. The amount of time one is in hold-down increases
exponentially with every flap.
11/28/2016
86
43
11/28/2016
Routing Information Protocol (RIP)

RIP is a simple interior gateway routing protocol
Built based on distance vector routing.
RIP is included as part of the UNIX software delivered with
most systems.
It is adequate for local area networks and is simple to
configure.
RIP selects the route with the lowest "hop count" ( metric )
as the best route
11/28/2016
87
An Example of RIP
Routers advertise the cost of
reaching networks.
In this example, Cs update to A
would indicate that C can reach
Networks 2 and 3 with cost 0,
Networks 5 and 6 with cost 1 and
Network 4 with cost 2.
4
A
11/28/2016
88
44
11/28/2016
Other RIP Details

Routing tables are exchanged every 30 seconds using the RIP
advertisement.
If a router does not hear from its neighbor once every 180
seconds, the neighbor is deemed unreachable.
The router that detects the failure will modify its routing
table and propagate the information.
11/28/2016
89
RIP implementation
RIP packets are sent using UDP.
Typically, there is a routing daemon (routed) that is an
application layer process that provides access to routing
tables.
Allows for the access of the tables.
The protocol is limited to networks whose longest path is 15
hops.
The designers believe that the basic protocol design is
inappropriate for larger networks.
11/28/2016
90
45
11/28/2016
RIP implementation
RIP uses numerous timers to regulate its performance.
These include a routing-update timer, a route timeout, and a routeflush timer.
The routing-update timer clocks the interval between periodic
routing updates. Generally, it is set to 30 seconds, with a small
random number of seconds added each time the timer is reset to
prevent collisions.
Each routing-table entry has a route-timeout timer associated with
it. When the route-timeout timer expires, the route is marked
invalid but is retained in the table until the route-flush timer
expires.
11/28/2016
91
Open Shortest Path First (OSPF)

Motivation
Original IGP used was RIP
Based on Bellman-Ford Algorithm
Worked well in small systems
Suffered from problems of Distance Vector Protocol
Large update packet packets
Slow response to topological changes
Slow Convergence
Need for a Link State Protocol

11/28/2016
92
46
11/28/2016
Open Shortest Path First

OSPF is based on the link-state routing protocol
OSPF is the most widely used IGP
A hierarchy is added -- OSPF runs within a domain or
administrative region. The region is sub-divided into areas.
Router does not need to know how to get to every network
within the domain -- enough to know how to get to right
area.
Load balancing - if two paths have same cost, subdivide
traffic among paths.
11/28/2016
93
OSPF Areas
Network is subdivided into areas
One special area called backbone area.
(Area 0).
Area 3
Area 1
Area 0
R9
R7
R8
R3
R1
R4
R2
Area 2
Within each area -- link state routing.

Link state advertisements of non border
routers do not leave area.
Packet goes from non-backbone area to
backbone area and crosses the backbone
into the Internet.
R6
R5
A router that is a member of both

the backbone and a non-backbone
area (R1, R2, R3) is called a area
router.
47
11/28/2016
Areas (continued)
Border routers summarize routing information and make it
available to other areas -- act like proxies --reflect costs to
reach networks from an area.
When there are many possible routes, routers choose cost
info to forward packets.
As a result of the hierarchy, OSPF can scale to handle much
larger AS than other routing protocols
Trade-offs -- Optimality versus scalability
All packet have to pass through the backbone area (may not be
optimal).
OSPF metrics
How does one assign link costs ?
Costs distinguish between different physical links.
Count the number of queued packets --does not take
bandwidth or latency into consideration.
Latency on the link.
Each incoming packet time-stamped. There is an associated
transmission time and propagation latency -- static for the
link.
Delay = (Depart time - Arrival time) + transmission time +
prop. latency.
11/28/2016
96
48
11/28/2016
OSPF Hello Message

Hello messages are used to allow a router to discover other adjacent
routers on its local links and networks.
During normal operation, routers send hello messages to their
neighbors at regular intervals (the hello interval).
If a router stops receiving hello messages from a neighbor, after a set
period (the dead interval) the router will assume the neighbor has
gone down.
These intervals must be the same on all routing devices on a shared
network. By default, the routing device sends hello packets every 10
seconds (broadcast and point-to-point networks) and 30 seconds
(nonbroadcast multiple access (NBMA) networks). By default, the dead
interval is four times the default hello interval, which is 40 seconds
(broadcast and point-to-point networks) and 120 seconds (NBMA
networks).
11/28/2016
97
Further reading
To understand more about a specific routing protocol, you
should read its RFC (Request for Comments), a formal
document from the Internet Engineering Task Force ( IETF )
that is the result of committee drafting and subsequent
review by interested parties.
https://www.ietf.org/rfc.html
11/28/2016
98
49

Chapter 4 Full

Transféré par

Informations du document

Description originale:

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Chapter 4 Full

Transféré par

Droits d'auteur :

Formats disponibles

11/28/2016

DR. PHUONG NGA TRAN - EEIT

DR. PHUONG NGA TRAN - EEIT

No routing problem as long as traffic intensities (load) are

DR. PHUONG NGA TRAN - EEIT

For datagram operation the packet contains the full

destination address in the packet header.

The routing table:

Routing table for node 4

DR. PHUONG NGA TRAN - EEIT

DR. PHUONG NGA TRAN - EEIT

Information used in the algorithm

DR. PHUONG NGA TRAN - EEIT

upon major load change

DR. PHUONG NGA TRAN - EEIT

Algorithms to determine shortest paths

The costs are then condensed in a cost matrix obeying the

DR. PHUONG NGA TRAN - EEIT

Algorithms to determine shortest paths

DR. PHUONG NGA TRAN - EEIT

Algorithms to determine shortest paths

All algorithms make use of Bellmann's optimality principle:

DR. PHUONG NGA TRAN - EEIT

Algorithms to determine shortest paths

be the minimal distance from node 1 to node .

be the minimal distance to node with j being a direct

( ) is the set of neighbours of .

DR. PHUONG NGA TRAN - EEIT

DR. PHUONG NGA TRAN - EEIT

DR. PHUONG NGA TRAN - EEIT

DR. PHUONG NGA TRAN - EEIT

DR. PHUONG NGA TRAN - EEIT

DR. PHUONG NGA TRAN - EEIT

DR. PHUONG NGA TRAN - EEIT

Summary of Dijkstra's Algorithm

DR. PHUONG NGA TRAN - EEIT

Pseudo code of Dijkstra's Algorithm

DR. PHUONG NGA TRAN - EEIT

DR. PHUONG NGA TRAN - EEIT

M = set of considered nodes

DR. PHUONG NGA TRAN - EEIT

Bellman Ford Algorithm

Applied for directed graphs

DR. PHUONG NGA TRAN - EEIT

RELAX EDGES REPEATEDLY

for each edge ( , ) do

error graph contains negative cycle

DR. PHUONG NGA TRAN - EEIT

DR. PHUONG NGA TRAN - EEIT

DR. PHUONG NGA TRAN - EEIT

DR. PHUONG NGA TRAN - EEIT

DR. PHUONG NGA TRAN - EEIT

DR. PHUONG NGA TRAN - EEIT

DR. PHUONG NGA TRAN - EEIT

Proof of Bellman Ford Algorithm

We always have ( ) ( , ) and if we ever find

Assume that after k iterations,

If we ever achieve ( ) = ( , ), we cannot decrease ( ) since we have

DR. PHUONG NGA TRAN - EEIT

Proof of Bellman Ford Algorithm

after relaxing edge

because p is the shortest path.

(upper bound property)