Transcript Ch18 WAN Technologies and Dynamic Routing

Department of Engineering Science
ES465/CES 440, Intro. to Networking & Network Management
WAN Technologies & Dynamic Routing
http://www.sonoma.edu/users/k/kujoory
References
• “Computer Networks & Internet,” Douglas Comer, 6th ed, Pearson, 2014, Ch 18,
Textbook, 5th ed, slides by Lami Kaya ([email protected]) with some changes.
• “Computer Networks,” A. Tanenbaum, 5th ed., Prentice Hall, 2011, ISBN:
13:978013212695-3.
• “Computer & Communication Networks,” Nader F. Mir, 2nd ed, Prentice Hall, 2015, ISBN:
13: 9780133814743.
• “Data Communications Networking,” Behrouz A. Forouzan, 4th ed, Mc-Graw Hill, 2007
• “Data & Computer Communications,” W. Stallings, 7th ed., Prentice Hall, 2004.
• “Computer Networks: A Systems Approach," L. Peterson, B. Davie, 4th Ed., Morgan
Kaufmann 2007.
Ali Kujoory
5/31/2016
Not to be reproduced without permission
1
Topics Covered
•
•
•
•
•
•
•
•
•
•
•
•
•
•
18.1 Introduction
18.2 Large Spans & Wide Area Networks
18.3 Traditional WAN Architecture
18.4 Forming a WAN
18.5 Store & Forward Paradigm
18.6 Addressing in a WAN
18.7 Next-Hop Forwarding
18.8 Source Independence
18.9 Dynamic Routing Updates in a WAN
18.10 Default Routes
18.11 Forwarding Table Computation
18.12 Distributed Route Computation
18.13 Shortest Path Computation in a Graph
18.14 Routing Problems
Ali Kujoory
5/31/2016
Not to be reproduced without permission
2
18.1 Introduction
• This chapter
– considers the structure of a network that spans an arbitrarily
large area
– describes the basic components used to build a packet
switching system
– explains the fundamental concept of routing
– presents the two basic routing algorithms
– explains the advantages of each algorithm
– extends the discussion of routing to the Internet
– presents routing protocols that use the algorithms described
here
Ali Kujoory
5/31/2016
Not to be reproduced without permission
3
18.2 Large Spans & Wide Area Networks
• Networking technologies
can be classified
according to the distance
spanned:
– PAN spans a region near an
individual
– LAN spans a building or
campus
– MAN spans a large
metropolitan area
– WAN spans multiple cities or
countries
Ali Kujoory
5/31/2016
• Consider a company that
uses a satellite bridge to
connect LANs at two sites
– Should the network be
classified as a WAN or as an
extended LAN?
Not to be reproduced without permission
4
18.2 Large Spans & Wide Area Networks
• The key issue that
separates WAN
technologies from LAN
technologies is scalability
• A WAN must be able to
grow as needed to
connect many sites
– spread across large
geographic distances
• A technology is not
classified as a WAN
unless it can deliver
reasonable performance
for a large scale network
– A WAN does not merely
connect to many computers
at many sites
• it must provide sufficient
capacity to permit all
computers to communicate
• Thus, a satellite bridge
that connects a pair of
PCs & printers is merely
an extended LAN
Ali Kujoory
5/31/2016
Not to be reproduced without permission
5
18.3 Traditional WAN Architecture
• WAN designers chose to
create a special-purpose
hardware device that
could be placed at each
site
• A packet switch provides
– local connections for
computers at the site, &
– connections for data circuits
that lead to other sites
• A packet switch consists
of a small computer
system
Ali Kujoory
5/31/2016
– with a processor, memory,
& I / O devices used to send
& receive packets
• Early packet switches
were constructed from
conventional computers
– the packet switches used in
the highest-speed WANs
require special-purpose
hardware
• Fig. 18.1 illustrates the
internal packet switch
architecture
Not to be reproduced without permission
6
18.3 Traditional WAN Architecture
Ali Kujoory
5/31/2016
Not to be reproduced without permission
7
18.3 Traditional WAN Architecture
• Since the advent of LAN technology, most WANs
separate a packet switch into two parts:
– a Layer 2 switch that connects local computers
– a router that connects to other sites
• Part 4 of the text
– discusses Internet routers in detail, &
– explains how the concepts covered here apply to the Internet
• Communication with local computers can be separated
from transmission across a WAN
• Fig. 18.2 illustrates the separation
Ali Kujoory
5/31/2016
Not to be reproduced without permission
8
18.3 Traditional WAN Architecture
or Layer 3 Switch
& Internet
Ali Kujoory
5/31/2016
Not to be reproduced without permission
9
18.4 Forming a WAN
• A WAN can be formed by
interconnecting a set of
sites
• The exact details of the
interconnections depend on
– the data rate needed
– the distance spanned, &
– the delay
• Many WANs use leased
data circuits
– However, other forms are
also available such as
• microwave & satellite
channels
Ali Kujoory
5/31/2016
• A network designer must
choose a topology
– For a given set of sites, many
topologies are possible
• Fig. 18.3 illustrates a
possible way to interconnect
– a WAN does not need to be
symmetric in the
interconnections among
packet switches
– the capacity of each
connection can be chosen to
accommodate the expected
traffic & provide redundancy
in case of failure
Not to be reproduced without permission
10
18.4 Forming a WAN
Ali Kujoory
5/31/2016
Not to be reproduced without permission
11
18.5 Store & Forward Paradigm
• The goal of a WAN is to
allow as many computers
as possible to send
packets simultaneously
• The fundamental
paradigm used to achieve
simultaneous transmission
is known as store &
forward
• To perform store &
forward processing
– a packet switch buffers
packets in memory
Ali Kujoory
5/31/2016
• The store operation
occurs when a packet
arrives:
– I / O hardware in the switch
places a copy of the packet in
memory
• The forward operation
occurs once a packet has
arrived & is waiting in
memory. The processor
– examines the packet
– determines its destination, &
– sends the packet over the I /
O interface that leads to the
destination
Not to be reproduced without permission
12
18.6 Addressing in a WAN
• From the view of an
attached computer
– a traditional WAN network
operates similar to a LAN
– Each WAN technology
defines the exact frame
format a computer uses
when sending & receiving
data
– Each computer connected to
a WAN is assigned an
address
• WANs addresses follow a
key concept that is used in
the Internet: hierarchical
addressing
Ali Kujoory
5/31/2016
Net_ID
Host_ID
– Hierarchical addressing
divides each address into two
parts:
(site, computer at the site)
– In practice, instead of a
identifying a site, each packet
switch is assigned a unique #
• first part of an address
identifies a packet switch
• second part identifies a
specific computer
• Fig. 18.4 shows two-part
hierarchical addresses
assigned to computers
connected to a pair of packet
switches
Not to be reproduced without permission
13
18.6 Addressing in a WAN
Ali Kujoory
5/31/2016
Not to be reproduced without permission
14
18.6 Addressing in a WAN
• Fig. 18.4 shows each
address as a pair of
decimal integers
– A computer connected to port
6 on packet switch 2 is
assigned address [2, 6]
• In practice, an address is
represented as a single
binary value with some
bits of the binary value
• In Part 4 of the text, we’ll
show that each Internet
address consists of a
binary #, where
– a prefix (NET_ID) of the bits
identify a specific network in
the Internet
– the remainder of the bits
identify a computer attached
to the network (HOST_ID)
– used to represent a packet
switch, &
– others used to identify a
computer
Ali Kujoory
5/31/2016
Not to be reproduced without permission
15
18.7 Next-Hop Forwarding
• What is the importance of
hierarchical addressing?
• When a packet arrives
• To make the choice, a
packet switch
– a switch must choose an
outgoing path over which to
forward it
• If a packet is destined
– for a local computer
• the switch sends the packet
directly to the computer
– Otherwise
• the packet must be forwarded
over to another switch
Ali Kujoory
5/31/2016
– examines the destination
address in the packet &
– extracts the packet switch #
• If the # in the destination
address is identical to the
packet switch's own ID the
packet is intended for a
computer on the local packet
switch
• Otherwise, the packet is
intended for a computer on
another switch
• Algorithm 18.1 explains the
computation
Not to be reproduced without permission
16
18.7 Next-Hop Forwarding
Ali Kujoory
5/31/2016
Not to be reproduced without permission
17
Figure 22.1 Direct & Indirect Delivery (from Forouzan)
22.18
Ali Kujoory
5/31/2016
Not to be reproduced without permission
18
Figure 22.2 Route Method versus Next-hop Method
Full Route
Long!
Ali Kujoory
5/31/2016
Not to be reproduced without permission
Next hop Route
Short
22.19
19
Figure 22.3 Host-specific Versus Network-specific Method
Host Specific
Long table!
Network Specific
Short table
Used for default
• Network specific makes the routing table much simpler & more
efficient.
Ali Kujoory
5/31/2016
Not to be reproduced without permission
22.20
20
Figure 22.4 Default Method
Ali Kujoory
5/31/2016
Not to be reproduced without permission
22.21
21
Figure 22.5 Simplified Forwarding Module In Classless Address
In classless addressing, we need at least
four columns in a routing table.
Ali Kujoory
5/31/2016
Not to be reproduced without permission
22.22
22
Example 22.1
For Fig. 22.6, make a routing table for router R1.
Open a blank table with the following
columns:
1. Mask column list the total number of
1-bits to be used for ANDing with the
desired destination addresses.
2. Network address with lists the
destination network addresses.
3. Next Hop lists the network addresses
toward the destination & the
Interface can take care of it.
4. Interface list the interfaces of the
current switch for the next hop.
Filling up the entries is very easy.
• Enter the network address of each
network, its corresponding mask, &
port it is connected to in the 2nd,1st, &
4th columns, respectively.
• The last row (Any) is used for all
addresses not covered by the
networks.
Table 22.1
Routing table for
router R1 in
Figure 22.6
Ali Kujoory
Not to be reproduced without permission
5/31/2016
Figure 22.6
Configuration for
Example 22.1
22.23
23
Example 22.2
Given the routing table, show the forwarding process if a packet arrives at R1 in Fig.
22.6 with the destination address 180.70.65.140.
The router performs the
following steps:
• 1st mask (/26) is applied to the
destination address. The result
is 180.70.65.128, which does
not match the corresponding
network address.
• 2nd mask (/25) is applied to
the destination address. The
result is 180.70.65.128, which
matches the corresponding
network address. The nexthop address & the interface
number m0 are passed to
ARP for further processing.
Ali Kujoory
5/31/2016
Not to be reproduced without permission
22.24
24
Example 22.3
Given the routing table, show the forwarding process if a packet arrives at R1 in Figure
22.6 with the destination address 201.4.22.35.
The router performs the following
steps:
• 1st mask (/26) is applied to the
destination address. The result is
201.4.22.0, which does not match
the corresponding network
address.
• 2nd mask (/25) is applied to the
destination address which does
not match the corresponding
network address.
• 3rd mask (/24) is applied to the
destination address. The result is
201.4.22.0, which matches the
corresponding network address.
The destination address of the
packet & the interface number m3
are passed to ARP.
Ali Kujoory
5/31/2016
Not to be reproduced without permission
22.25
25
Example 22.4
Given the routing table, show the forwarding process if a packet arrives at R1 in
Figure 22.6 with the destination address 18.24.32.78.
This time all masks are
applied, one by one, to the
destination address, but no
matching network address is
found.
• When it reaches the end of
the table, the module gives
the next-hop address
180.70.65.200 & interface
number m2 to ARP.
• This is probably an outgoing
package that needs to be
sent, via the default router,
to someplace else in the
Internet.
Ali Kujoory
5/31/2016
Not to be reproduced without permission
22.26
26
18.7 Next-Hop Forwarding
• A packet switch does not need
to keep complete information
about how to reach all
possible computers
– nor does a switch need to
compute the entire route a
packet will follow
– which means that a switch only
needs to know which outgoing
link
• A switch only needs to compute
the next hop for a packet
5/31/2016
– is analogous to the way airlines
list flights
• To make the computation
efficient
• Instead, a switch bases
forwarding on packet switch IDs
Ali Kujoory
• The process is called next-hop
forwarding &
– switches use table lookup
– that is, each packet switch
contains a forwarding table
– such tables were originally called
routing tables
• lists all possible packet switches
• & gives a next hop for each
• Fig. 18.5 illustrates next-hop
forwarding with a trivial
example
Not to be reproduced without permission
27
18.7 Next-Hop Forwarding
Ali Kujoory
5/31/2016
Not to be reproduced without permission
28
18.7 Next-Hop Forwarding
• Using only one part of a twopart hierarchical address to
forward a packet has two
practical consequences
1. the computation time required
to forward a packet is reduced
because the forwarding table can
be organized as an array that
uses indexing instead of
searching
2. the forwarding table contains one
entry per packet switch instead
of one entry per destination
computer
• The reduction in table size can
be substantial
Ali Kujoory
5/31/2016
– Esp. for a large WAN that has
many computers attached to
each packet switch
• A two-part hierarchical
addressing scheme allows
packet switches to use only the
first part (Net_ID) of the
destination address until the
packet reaches the final switch
– Once the packet reaches the
final switch
• the switch uses the second part of
the address (Host-ID) to choose a
specific computer
– As Algorithm 18.1 described
Not to be reproduced without permission
29
18.8 Source Independence
• Next-hop forwarding does
not depend on the packet's
original source or on the
path the packet has taken
before it arrives at a
particular packet switch
– Instead, the next hop to which
a packet is sent depends only
on the packet's destination
– The concept is known as
source independence
• Like postal service & Airline
• Source independence allows
the forwarding mechanism in
a computer network to be
compact & efficient
Ali Kujoory
5/31/2016
– Because all packets follow
the same algorithm, only one
table is required
– Because forwarding does not
use source information, only
the destination address
needs to be extracted from a
packet
– A single mechanism handles
forwarding uniformly packets
that originate on directly
connected computers &
• packets that arrive from other
packet switches use the same
mechanism
Not to be reproduced without permission
30
Desirable Characteristics of Routing Algorithms
•
•
•
•
•
•
Correct to deliver
Simple to implement
Robust against failures
Stable against traffic variation (converge to equilibrium)
Fair to all users
Optimal to network
– least delay, max throughput, or least jitter
• Dynamic (Adaptive to changes) in
– Topology
• Failures of network elements
– Configuration
• Network elements added / removed
– Traffic
• Load of the network
• Static routing algorithms
o not adaptive to changes
o Are unacceptable for
large networks
• Dynamic routing algorithm
is used for large networks
since it is most efficient
– Quality of service
• Delay, throughput, loss
Ali Kujoory
5/31/2016
Not to be reproduced without permission
31
18.9 Dynamic Routing Updates in a WAN
Forwarding table must
guarantee the following:
• Universal communication
– The forwarding table in each
switch must contain a valid
next-hop route for each
possible destination address
• Optimal routes
– In a switch, the next-hop
value in the forwarding table
for a given destination must
point to the shortest path to
the destination
• Network failures further
complicate forwarding
Ali Kujoory
5/31/2016
– E.g., if two paths exist to a
given destination & one of the
paths becomes unavailable
because hardware fails,
forwarding should be
changed to avoid the
unavailable path
– A network manager cannot
merely configure a forwarding
table that contain static
values that do not change
– Instead, software running on
the packet switches
continually tests for failures, &
reconfigures the forwarding
tables automatically
Not to be reproduced without permission
32
18.9 Dynamic Routing Updates in a WAN
• We use the term routing
software to describe
software that automatically
reconfigures forwarding
tables
• Route computation in a
WAN is to think of a graph
that models the network
– software uses the graph to
compute the shortest path to
all possible destinations
• Each node in the graph
corresponds to a packet
switch in the network
– individual computers are not
part of the graph
• If the network contains a
direct connection between a
pair of packet switches
– the graph contains an edge
or link between the
corresponding nodes
• Fig. 18.6 shows an example
WAN & the corresponding
graph
Ali Kujoory
5/31/2016
Not to be reproduced without permission
33
18.9 Dynamic Routing Updates in a WAN
• For dynamic routing computation, graphs are used since
mathematical tools are available
Ali Kujoory
5/31/2016
Not to be reproduced without permission
34
18.9 Dynamic Routing Updates in a WAN
• As Fig. 18.6 shows, nodes in
the graph are given a label
– is the same as the # assigned
to the corresponding packet
switch
• A graph representation is
useful in computing next-hop
forwarding
– because graph theory has
been studied & efficient
algorithms have been
developed
– a graph abstracts away
details, allowing routing
software to deal with the
essence of the problem
Ali Kujoory
5/31/2016
• When it computes next-hop
forwarding for a graph
– a routing algorithm must
identify a link
• Our examples will use the
notation (k, j) to denote a
link from node k to node j
– When a routing algorithm
runs on the graph in Fig.
18.6b
– The algorithm produces
output as shown in Fig. 18.7
Not to be reproduced without permission
35
18.9 Dynamic Routing Updates in a WAN
Fig. 18.6b graph of WAN.
Fig. 18.7 A forwarding table for each node in the graph of Fig. 18.6b.
Ali Kujoory
5/31/2016
Not to be reproduced without permission
36
18.10 Default Routes
• The forwarding table for
node 1 in Fig. 18.7 raises
an important point:
– a forwarding table may
contain many entries that
point to the same next hop
• An examination of the
WAN in Fig. 18.6a reveals
why all remote entries
contain the same next
hop:
– the packet switch has only
one connection to the
network
Ali Kujoory
5/31/2016
– therefore, all outgoing traffic
must be sent across the
same connection
– consequently, except for the
entry that corresponds to the
node itself,
• all entries in node1's
forwarding table have a next
hop that points to the link from
node1 to node3
• In our trivial example, the
list of duplicate entries in
the forwarding table is
short
Not to be reproduced without permission
37
18.10 Default Routes
• A large WAN may contains
hundreds of duplicate entries
• Most WAN systems include
a mechanism that can be
used to eliminate the
common case of duplicate
entries
• Default route is a
mechanism that allows a
single entry in a forwarding
table to replace a long list of
entries that have the same
next-hop value
– & the entry has lower priority
than other entries
• If the forwarding mechanism
does not find an explicit
entry for a given destination
– it uses the default
• Fig. 18.8 shows the
forwarding tables from Fig.
18.7 revised to use a default
route
– Only one default entry is
allowed in a forwarding table
Ali Kujoory
5/31/2016
Not to be reproduced without permission
38
18.10 Default Routes
Fig. 18.7 A forwarding table
for each node in the graph of
Fig. 18.6b.
* = Default Route makes
the table shorter
Fig. 18.8 A forwarding tables from Fig. 18.7 with default routes denoted by an an asterisk.
Ali Kujoory
5/31/2016
Not to be reproduced without permission
39
18.10 Default Routes
• Default routing is optional
– a default entry is present only if more than one
destination has the same next-hop value
• E.g., the forwarding table for node3 does not
contain a default route
– because each entry has a unique next hop
• However, the forwarding table for node1 benefits
from a default route
– because all remote destinations have the same next
hop
Ali Kujoory
5/31/2016
Not to be reproduced without permission
40
18.11 Forwarding Table Computation
Basic approaches to construct
forwarding tables:
• Static routing
– A program computes &
installs routes when a packet
switch boots
– The routes do not change
• Dynamic routing
– A program builds an initial
forwarding table when a
switch boots
– Program then alters the table
as conditions in the network
change
Ali Kujoory
5/31/2016
• Each approach has
advantages & disadvantages
– Advantages of static routing
are simplicity & low overhead
– Disadvantage is inflexibility
• static routes cannot be
changed when
communication is disrupted
• Large networks are
designed with redundant
connections to handle
occasional hardware failures
– most WANs use a form of
dynamic routing
Not to be reproduced without permission
41
Figure 22.12 Autonomous Systems
• An internet can be so large that one routing
protocol cannot handle updating the routing
tables.
• Thus, an internet is divided into
Autonomous System.
• An AS is a group of networks &
– routers are under the authority of a single
administration.
22.42
Ali Kujoory
5/31/2016
• Routing inside an autonomous system is
called intradomain routing.
– Each AS can choose one or more intradomain
routing protocols inside AS.
• Routing outside an autonomous system is
called interdomain routing.
– Only one interdomain routing protocol can be used
between ASes
Not to be reproduced without permission
42
Figure 22.13 Popular Routing Protocols
(Routing Information
Protocol)
22.43
Ali Kujoory
5/31/2016
(Open Shortest Path
First)
(Border Gateway
Protocol)
Not to be reproduced without permission
43
Figure 22.14 Distance Vector Routing (DVR) Tables
• In DVR, the least-cost route between
any two nodes is the route with
minimum distance.
• Each node maintains a vector (table)
of minimum distance to every node.
• The table at each node guides the
packets to the desired node using the
next hop in the route, e.g.,
– In an area, cities are nodes, lines are the roads
connecting the cities we take to destination.
• DVR uses three phases: Initialization,
Sharing, Updating.
Cost is a metric,
e.g., # of hops
22.44
Ali Kujoory
5/31/2016
Not to be reproduced without permission
44
Figure 22.15 Initialization of Tables In Distance Vector Routing
Initialization:
• In the beginning, each node can know only the distance between
itself & its immediate neighbors, those directly connected to it.
• Use infinity for the distance (cost) to the nodes not neighbor (not
known).
22.45
Ali Kujoory
5/31/2016
Not to be reproduced without permission
45
Sharing of Tables with all Neighboring Nodes Routing
In distance vector routing, each node shares its routing
table with its immediate neighbors periodically &
when there is a change.
Sharing:
• Each node shares its routing table with its neighboring node, e.g.,
– Although node A does not know about node E, node C does.
• So if node C shares its routing table with A,
– node A can also know how to reach E, & so on.
• Question:
– How much of the table should be shared?
– The 3rd column does not help, so it is not sent.
22.46
Ali Kujoory
5/31/2016
Not to be reproduced without permission
46
Figure 22.16 Updating in Distance Vector Routing
Updating in three steps:
1. Receiving node adds the cost between itself & sending node in the 2nd
column.
 If node C claims its distance to a destination is x mi, & the distance between A & C
is y mi, the distance between A & that destination, via C, is x + y mi,
2. Receiving node adds the name of the sending node to the 3rd column for all
entries indicating the sending node is the next node in the route.
3. Receiving node compares each row of its old table with the corresponding
row of the modified version of the received table & uses the lower cost.
22.47
Ali Kujoory
5/31/2016
Not to be reproduced without permission
47
Routing Information Protocol (RIP)
• RIP is an intradomain routing protocol used in an autonomous
system.
• It is very simple routing protocol based on DVR.
• The destinations in a routing table is a network, thus the first
column defines a network address.
• RIP uses hop count for the metric, that makes things very
simple.
• Infinity is defined as 16; so any route used by RIP in AS
cannot be more than 15.
• The next node column defines the address of the router in
which the packet is to be sent to reach the destination.
22.48
Ali Kujoory
5/31/2016
Not to be reproduced without permission
48
Figure 22.19 Example of a Domain Using RIP
22.49
Ali Kujoory
5/31/2016
Not to be reproduced without permission
49
Figure 22.20 Link State Routing (LSR)
• In LSR each node has the entire
topology of the domain
– The list of nodes & links, how they
are connected for cost (metric), &
– whether they are up or down
• Nodes use Dijkstra’s algorithm to
build the routing table
https://en.wikipedia.org/wiki/Dijkstra%27s_algorithm
• Analogy: In a city, each person
may the same map but uses
different route to reach her/his
destination.
Domain, same
for all nodes
Routing table, different
for each node
22.50
Ali Kujoory
5/31/2016
Not to be reproduced without permission
50
Figure 22.21 Link State Knowledge
22.51
Ali Kujoory
5/31/2016
Not to be reproduced without permission
51
How Link-State Works https://en.wikipedia.org/wiki/Link-state_routing_protocol
• The link-state protocol is
performed by every switching node
(routers that forward packets) in
the network
• Every node constructs a map of the
connectivity to the network
– in the form of a graph, showing
– which nodes are connected to which
other nodes
• Each node then independently
calculates the next best logical
path from it to every possible
destination in the network
• The collection of best paths will
then form the node's routing table
Ali Kujoory
5/31/2016
• This contrasts with DistanceVector Routing protocols which
work by having each node share its
routing table with its neighbors,
• In a link-state protocol the only
information passed between nodes
is connectivity related
• Each node floods its LSP to every
other router in an efficient &
reliable way
– Link-state algorithms are sometimes
characterized informally as each
router “telling the world about its
neighbors”
Not to be reproduced without permission
52
Shortest Path Routing Example
Problem
• Find shortest path between
nodes A & D
• Use Dijkstra’s algorithm
Solution
• Build a graph of the subnet with
– Each node representing a router
– Each arch representing a link
• Make a weighted undirected
graph
– By marking each arc with the
desired path metric
– Examples of path metric:
• # Of hops, geographic distance,
queuing / transmission delay, BW,
communication costs
Ali Kujoory
5/31/2016
• In the figures of the following
slide
– Weights on the arcs show the
distances = path metric
– Arrow indicates the working node
– (¥,-) indicates path is not known
• Start with node A
• Then examine each adjacent
node
• Relabel each by distance from
source node along the best
known path
• Continue this to other nodes
See also the program in Tanenbaum
Fig. 5-8, oo=
¥
Not to be reproduced without permission
53
Shortest Path Routing (2)
B
2
7
2
A
6
E
2
C
F
1
3
F(oo,-)
G(6,A)
B(2,A)
F(6,E)
D(oo,-)
G(5,E)
H(9,G)
(e) F is working node. It is made permanent.
Ali Kujoory
5/31/2016
F(6,E)
D(oo,-)
H(oo,-)
G(5,E)
(d) Shorter distance from A is via E. So
relabel & mark permanent.
B(2,A)
C(9,B)
C(9,B)
E(4,B)
D(oo,-)
C(9,B)
E(4,B)
D(oo,-) A
(c) Calculate distances from A. If shortest, label
the node as permanent.
A
F(oo,-)
G(6,A)
H(oo,-)
(b) Relabel adjacent nodes with distance from A.
H(oo,-)
B(2,A)
E(oo,-)
A
C(9,B)
E(4,B)
A
C(oo,-)
2
4
G
H
(a) Start at node A & mark it as permanent.
B(2,A)
B(2,A)
3
D
2
via
E(4,B)
A
F(6,E)
G(5,E)
D(10,H)
H(8,F)
(f) H is working node. Via F is shorter.
Relabel & make permanent.
Not to be reproduced without permission
54
18.12 Distributed Route Computation
• Algorithm 18.2 shows how a
forwarding table can be
computed
• There are two general forms:
– after information about a
network is encoded in a
graph
– Link-State Routing (LSR),
which uses Dijkstra's
algorithm
– Distance-Vector Routing
(DVR), which uses another
approach
• In practice, WANs need to
perform distributed route
computation
– Instead of a centralized
program computing all
shortest paths
• The next sections describe
each of the two approaches
• each packet switch must
compute its own forwarding
table locally
Ali Kujoory
5/31/2016
– All packet switches must
participate in distributed
route computation
– Chapter 26 explains how
each approach is used to
control routes in the Internet
Not to be reproduced without permission
55
18.12.1 Link-State Routing (LSR)
• Link-state routing or Linkstatus routing
– the approach also known as
Shortest Path First (SPF)
routing
– Dijkstra algorithm used it to
characterize the way it works
• actually all routing algorithms
find shortest paths
• To use LSR, packet
switches periodically send
messages across the
network that carry the status
of a link
– Link-State
Ali Kujoory
5/31/2016
– E.g., packet switches 5 & 9
measure the link between
them & send a status
message
• such as “the link between 5 &
9 is up”
– Each status message is
broadcast to all switches
• Every switch collects
incoming status messages
– & uses them to build a graph
of the network
• Each switch then uses
Algorithm 18.2 to produce a
forwarding table by choosing
itself as the source
Not to be reproduced without permission
56
Algorithm 18.2
A version of Dijkstra’s
algorithm that computes R,
a nexthop forwarding table,
& D, the distance to each
node from the specified
source node
Ali Kujoory
5/31/2016
Not to be reproduced without permission
57
18.12.1 Link-State Routing (LSR)
• An LSR algorithm can adapt to hardware failures
• If a link between packet switches fails
– the attached packet switches will detect the failure &
– broadcast a status message that specifies the link is down
• All packet switches receive the broadcast
– change their copy of the graph
• to reflect the change in the link's status &
– re-compute shortest paths
• Similarly, when a link becomes available again
– the packet switches connected to the link detect that
• it is working &
– start sending status messages that report its availability
Ali Kujoory
5/31/2016
Not to be reproduced without permission
58
18.12.2 Distance Vector Routing (DVR)
• In DVR, each link in the
network is assigned a
weight, as with LSR
• The distance to a
destination between two
packet switches is defined
to be the sum of weights
along the path between
the two
• Like LSR, DVR arranges
for packet switches to
exchange messages
periodically
Ali Kujoory
5/31/2016
• In DVR, a switch sends a
complete list of
destinations & the current
cost of reaching each
• When it sends a DVR
message
– a switch is sending a series of
individual statements, of the
form:
“I can reach destination X, &
its current distance from me
is Y”
Not to be reproduced without permission
59
18.12.2 Distance Vector Routing (DVR)
• DVR messages are not
broadcast
– Each switch periodically
sends a DVR message to its
neighbors
• Each message contains
pairs of (destination, distance)
• Each packet switch must
keep a list 0f
• DVR software can be
considered as maintaining
an extension to the
forwarding table that
– stores a distance for each
destination
– possible destinations,
– the current distance to the
destination,
– the next hop to use for each
in the forwarding table
Ali Kujoory
5/31/2016
Not to be reproduced without permission
60
18.12.2 Distance Vector Routing (DVR)
• When a message arrives at a switch from neighbor N
– the switch examines each item in the message &
– changes its forwarding table if the neighbor has a shorter path to
some destination than the path currently being used
Example:
• If neighbor N advertises a path to destination D as having
cost 5 & the current path through neighbor K has cost
100
– the current next hop for D will be replaced by N &
– the cost to reach D will be 5 plus the cost to reach N
• Algorithm 18.3 specifies how routes are updated when
using the distance-vector approach
Ali Kujoory
5/31/2016
Not to be reproduced without permission
61
Algorithm 18.3
Distance-vector algorithm
for route computation
Ali Kujoory
5/31/2016
Not to be reproduced without permission
62
18.13 Shortest Paths & Weights
• Once a graph has been created that corresponds to a
network
– software uses a method known as Dijkstra's Algorithm
• To find the shortest path from a source node to each of
the other nodes in the graph:
– A next-hop forwarding table is constructed during the computation
of shortest paths
– The algorithm must be run once for each node in the graph
– That is, to compute the forwarding table for packet switch P
• the node that corresponds to P is designated as the source node &
• the algorithm is run
Ali Kujoory
5/31/2016
Not to be reproduced without permission
63
18.13 Shortest Paths & Weights
• Dijkstra's algorithm is popular
– because it can be used with various definitions of shortest path
– In particular, the algorithm does not require edges in the graph to
represent geographic distance. Instead, the algorithm
• allows each edge to be assigned a nonnegative value called a weight
• defines the distance between two nodes to be the sum of the weights
along a path between the nodes
• Fig. 18.9 illustrates the concept of weights by
– showing an example graph with an integer weight assigned to
each edge &
– a least-weight path between two nodes in the graph
Ali Kujoory
5/31/2016
Not to be reproduced without permission
64
18.13 Shortest Paths & Weights
Ali Kujoory
5/31/2016
Not to be reproduced without permission
65
18.13 Shortest Paths & Weights
• Dijkstra's algorithm
maintains a Set of nodes, S
– for which the min distance &
next hop have not been
computed
• The set is initialized to all
nodes except the source
• The algorithm then iterates
until set S is empty
• On each iteration
– the algorithm removes a node
from S that has the least
distance from the source
– As it deletes node u, the
algorithm examines the
current distance from the
source to each of the
neighbors of u
• that remains in the set
– If a path from the source
through u to the neighbor has
less weight than the current
path, the algorithm updates the
distance to the neighbor
• After all nodes have been
removed from S
– the algorithm will have
computed the min distance to
each node &
• a correct next-hop forwarding
table for all possible paths
Ali Kujoory
5/31/2016
Not to be reproduced without permission
66
18.13 Shortest Paths & Weights
• Dijkstra's algorithm needs 3
data structures to store:
– current distance to each node
– next hop for the shortest path
– information about the
remaining set of nodes
• Nodes can be numbered
from 1 to n as Fig. 18.9
– which makes the
implementation efficient
• because a node # can be used
as an index into a data
structure
• The algorithm can use two
arrays, D & R
– each indexed by the node #
– The ith entry in array D stores
a current value of the
minimum distance from the
source to node i
– The ith entry in array R stores
the next hop used to reach
node i along the path being
computed
– The set S can be maintained
as a doubly linked list of
node #s
• which facilitates searching the
entire set or deleting an entry
Ali Kujoory
5/31/2016
Not to be reproduced without permission
67
18.13 Shortest Paths & Weights
• Algorithm 18.2 uses
weight (i, j) as a function
– that returns the weight of the
edge from node i to node j
• Function weight is
assumed to return a
reserved value infinity if
no edge exists from node i
to node j
– In practice, any value can be
used to represent infinity
provided the value is larger
than the sum of weights along
any path in the graph
Ali Kujoory
5/31/2016
– One way to generate a value
infinity consists of adding one
to the sum of all weights on
all edges
• Allowing arbitrary weights
to be assigned to edges of
a graph means that
– one algorithm can be used
with different measures of
distance
– E.g., some WAN technologies
measure distance by
counting the # of packet
switches along a path
Not to be reproduced without permission
68
18.13 Shortest Paths & Weights
• To use the algorithm for
such technologies
• the other designated to be a
backup path
• To enforce such a policy
– each edge in the graph is
assigned a weight of 1
• In other WAN technologies,
weights are assigned to
reflect the capacity of the
underlying connections
– A network manager can
assign weights to links to
control routing
– E.g., consider a case where
two separate paths exist
between a pair of packet
switches with
– a network manager can
assign the primary link a low
weight & the other link a high
weight
– Routing software will
configure forwarding tables to
use the path with low weight
• unless the path is not available
• in which case routing software
will select the alternative path
• one path designated to be the
primary path &
Ali Kujoory
5/31/2016
Not to be reproduced without permission
69
18.14 Routing Problems
• In theory, either LSR or
DVR routing will compute
shortest paths
• Furthermore, each
approach will eventually
converge
– meaning that the forwarding
tables in all packet switches
agree
• However, problems do
occur
– E.g., if LSR messages are
lost, two packet switches can
disagree about the shortest
path
Ali Kujoory
5/31/2016
• DVR problems can be
more severe
– because a link failure
can cause two or more
packet switches to
create a routing loop
• in which each packet switch
thinks the next packet switch
in the set is the shortest path
to a particular destination
• As a result, a packet can
circulate among the switches
indefinitely
Not to be reproduced without permission
70
The Count-to-Infinity Problem
• Distance Vector Routing
converges slowly
• Routing loops can occur when
there is invalid update due to a
failure
• Then A comes up
– After 1st exchange B learns A is 1
hop away; others think A is still down
– After 2nd exchange, C learns A is 2
hops away; D & E think A is down
– In this case, the Distance-Vector is
incremented to a big #, Count-to-infinity
• In (a), initially A is down, other
routers know it (delay to A=infinity)
• Need N exchanges for a path of N
hops
• In (b) A is initially up & then goes
down; Count-to-infinity occurs
The delay metric = number of hops
After A comes back B is one hop from A.
B thinks it can go to A via C!
Ali Kujoory
5/31/2016
Not to be reproduced without permission
71
18.14 Routing Problems
• One of the primary reasons
DVR protocols exhibit problems
comes from backwash, i.e.,
– a packet switch receives
information that it sent
• E.g., suppose a switch tells its
neighbors
“I can reach destination D at cost
3”
– If the connection leading to
destination D fails
• Imagine that just after the link
fails
– one of the neighbors sends a
DVR message that specifies
“I can reach destination D at cost
4”
• Unfortunately
– the message will be believed &
– a routing loop will be created
• the switch will remove the entry for
D from its forwarding table, or
⌐ mark the entry invalid
– But the switch has told neighbors
that a route exists
Ali Kujoory
5/31/2016
Not to be reproduced without permission
72
18.14 Routing Problems
• Most practical routing mechanisms contain constraints &
heuristics (by try & error) to prevent problems like
routing loops
• E.g., DVR schemes employ split horizon
– which specifies that a switch does not send information back to its
origin
• Furthermore, most practical routing systems introduce
hysteresis (lag in effect)
– that prevents the software from making many changes in a short
time
• However, in a large network where many links fail &
recover frequently, routing problems can occur
Ali Kujoory
5/31/2016
Not to be reproduced without permission
73
Distance-vector versus Link-state
Distance Vector
Link-State
Views net topology from
Gets common view of entire
neighbor’s perspective
network topology
Adds distance vectors from
router to router
Frequent, periodic updates
- slow convergence
Passes copies of routing table
to neighbor routers
Calculates the shortest path to other
routers
Event-triggered updates
- faster convergence
Passes link-state routing updates to
other routers
Requires relatively less memory
Requires more memory / processing
/ processing
for handling routing tables
Used for smaller to medium size
Used for very large network (e.g.,
networks
Ali Kujoory
Internet) - cheaper
5/31/2016
Not to be reproduced without permission
74