Interior Gateway Routing Internet Service Providers Forum March 6, 1998

1
Interior Gateway RoutingInternet Service
Providers ForumMarch 6, 1998
by Patrick W. Gilmore pgilmore_at_pgexch.com,
patrick_at_ianai.net
2
Agenda

• Introduction, Requirements and Assumptions
• What is an Interior Gateway Protocol?
• IGP vs. EGP
• Different types of IGPs
• Static Routing
• RIP
• OSPF

3
Agenda

• EIGRP
• iBGP
• Interactions with BGP
• Examples of Each Protocol
• Multiple Protocol Networks
• Questions

4
Introductions, Requirements and Assumptions
5
Introduction

• This is a class on Interior Gateway Routing
  protocols, commonly called IGPs.
• It is designed for small to medium sized Internet
  Service Providers.

6
Introduction

• This class covers
• Static Routing
• RIP
• OSPF
• EIGRP
• iBGP

7
Introduction

• This class does not cover IS-IS or IGRP.
• This class does cover interaction with BGP, but
  does not explain how to configure BGP in depth.

8
Requirements

• This class is a beginners class for IGPs, but it
  is not a beginners routing class. Everyone here
  should be familiar with the general rules of IP
  routing and CIDR.

9
Assumptions

• All routing examples and configurations are using
  Cisco Systems routers. The default behavior and
  configuration of other vendors hardware may
  differ.
• Most of the concepts discussed are defined by
  standards (RFCs, etc.) and should apply to all
  routes regardless of vendor. (Except EIGRP,
  which is cisco proprietary.)

10
What Is anInterior Gateway Protocol?
11
Interior Gateway Protocols

• IGPs are designed to exchange network and
  subnetwork prefix information among routers
  within the same autonomous system--that is, among
  routers running a common routing protocol under
  on administrative domain. (Cisco Router
  Configuration, page 157)

12
What is an IGP?

• In English, that means an IGP is designed to let
  all of YOUR routers talk to each other.
• To put it another way, you would not use an IGP
  to trade routes with another network.

13
What is an IGP?

• Questions

14
IGP vs. EGP
15
Exterior Gateway Protocol

• EGP stands for Exterior Gateway Protocol.
• EGPs are used to exchange routes between
  different autonomous systems - different
  administrative domains.

16
IGP vs. EGP

• But why would we have different protocols for
  internal and external communications?
• This is mostly because of the different amounts
  of information.

17
IGP vs. EGP

• There are many more routes external to your
  network than there are internal.
• You also have much greater control over internal
  route selection and metrics than external.

18
IGP vs. EGP

• EGPs are more scalable than IGPs.
• IGP route selection is not more elegant and
  usually more reliable than EGP.
• You wouldnt want an IGP doing EGP work - trust
  me.

19
IGP vs. EGP

• Questions

20
Different Types of IGPs
21
Different types of IGPs

• There are seven types of IGPs commonly in use for
  IP routing today. These are

• Static
• RIP
• OSPF
• IS-IS

• IGRP
• EIGRP
• iBGP

22
Different types of IGPs

• These are categorized into two major categories
• Static
• Dynamic

23
Static Routes

• Static routes are well, static. They are
  defined by the user and never change. By default,
  they take precedence over dynamically learned
  routes.

24
Static Routes

• An example of a static route is
• ip route 0.0.0.0 0.0.0.0 10.1.1.1
• (In cisco-speak)

25
Static Routes

• ip route 0.0.0.0 0.0.0.0 10.1.1.1
• This command would force the router to send all
  default traffic to 10.1.1.1 no matter what
  route was learned through any of the other
  routing protocols.

26
Dynamic Routes

• Dynamic routes are learned from other routers.
  They can change without human intervention.
• All protocols other than static routes are
  dynamic.

27
Dynamic Routes

• Of the dynamic routing protocols, we will not
  discuss IS-IS or IGRP.

28
Different types of IGPs

• Of the dynamic routing protocols, there are two
  sub-types
• Classful
• Classless

29
Classful vs. Classless

• There are only two Classful routing protocols
  still in use RIPv1 and IGRP.
• Fortunately, we will not discuss IGRP. We will
  discuss RIPv1, but only briefly.

30
Different types of IGPs

• Of the dynamic, classless routing protocols,
  there are two sub-types
• Link-State
• Distance-Vector

31
Link-State Protocols

• OSPF and IS-IS are the only link-state protocols
  currently in use.

32
Distance-Vector Protocols

• EIGRP, IGRP, RIP and iBGP are all distance-vector
  protocols.
• (Actually, EIGRP uses a cisco proprietary
  algorithm called DUAL, but it is essentially a
  DV protocol.)

33
Interior Gateway Protocols

• Questions

34
Static Routing
35
Static Routing

• Static Routing is the most commonly used interior
  gateway protocol.

36
Static Routing

• Some reasons static routing is so popular
• Simple to Implement
• All routers support it
• Easy to troubleshoot
• Low CPU requirements/usage
• Low bandwidth requirements
• Does not break often

37
Static Routing

• There are also situations in which dynamic
  routing is not possible, or at least not
  desirable.
• The simplest case of this is a single-homed
  provider.

38
Static Routing

• Most upstreams will not provide BGP for
  single-homed downstreams. It is a waste of
  resources to do so. But the downstream needs to
  know where to send packets.
• The downstream can install a static default route
  and point it at the upstream.

39
Static Routing

• What is more, the upstream will likely install a
  static route for the downstreams prefixes and
  point it at the downstream.
• The upstream gains the same benefits from using
  static routing as the downstream.

40
Static Routing

• Questions

41
RIP
42
RIP

• RIP is the Routing Information Protocol.
• It comes in two versions called, aptly enough,
  version 1 and version 2.

43
RIP

• RIPv1 is defined in RFC 1058.
• RIPv2 is defined in RFC 1723.

44
RIP

• RIPs primary saving grace is that it is the
  oldest of all dynamic protocols still in use.
  This means that essentially every routing device
  available supports RIP. This is not true of
  other protocols (although OSPF is coming close).

45
RIP - Distance Vector

• RIP is a distance-vector based routing protocol.
  The vector in RIP is router hops.
• RIP inserts prefixes into the routing table with
  a hop count. The prefix with the lowest hop
  count is chosen.

46
RIP - Distance Vector

• A router running RIP will broadcast its entire
  routing table out each interface every 30
  seconds.
• Adjacent routers will hear this update, add one
  hop and calculate the best path to each
  destination.

47
RIP - Distance Vector

• RIP is the only protocol still in use which
  re-broadcasts routes on every advertisements.
  All other protocols send the full table on
  startup and changes only after that.
• Sending the full table can waste significant
  bandwidth, especially on WAN links.

48
RIP - Distance Vector

• Remember, the number of hops is the only metric
  advertised with each prefix. The route selection
  algorithm is based solely on how many hops to
  each destination.
• Because router hops are not all created equal,
  RIP has several problems scaling to large or
  complicated networks.

49
RIP - Distance Vector

• Consider the following network of four routers.
  There are three T1s and a 56Kbps backup link.

Router 2
Router 1
T1s
56K
Router 3
Router 4
50
RIP - Distance Vector

• It is obvious that to get from Router 1 to Router
  4, one should traverse Routers 2 and 3, but RIP
  would pick the 56Kbps link because of the lower
  hop count.

Router 2
Router 1
T1s
56K
Router 3
Router 4
51
RIP - Convergence

• RIP also has a problem with convergence.
• When a link breaks, RIP does not inform its peers
  of the change. The route announcement is simply
  not advertised.

52
RIP - Convergence

• Because of this, adjacent routers have to wait
  until the advertisement times out. By default
  this is 3 times the advertisement frequency, or
  90 seconds.

53
RIP - Convergence

• Using our previous example, assume the link
  between Router 1 and Router 2 dies.

Router 2
Router 1
T1s
56K
Router 3
Router 4
54
RIP - Convergence

• If Router 3 was sending data to Router 1 via
  Router 2, it would take 90 seconds for Router 3
  to send the data to router 4.

Router 2
Router 1
T1s
56K
Router 3
Router 4
55
RIP - Convergence

• Unfortunately, Router 2, knowing that the link to
  Router 1 is down, and hearing a route from Router
  3, sends the data back to Router 3.

Router 2
Router 1
T1s
56K
Router 3
Router 4
56
RIP - Convergence

• This, of course, causes a routing loop,
  potentially congesting the R1/R2 link so much
  even legitimate traffic will not get through.

Router 2
Router 1
T1s
56K
Router 3
Router 4
57
RIP - Convergence

• But wait, it gets worse. Not only can traffic
  loop, but a routing loop can be caused.
• Assume a simple linear network

58
RIP - Convergence

• Further assume R1 dies. R2 take 90 seconds to
  time out the R1 routes, during which time it will
  continue to advertise these routes to R3.

Router 1
Router 3
Router 2
59
RIP - Convergence

• At which time, R3 will stop seeing R1 routes from
  R2. However, R3 will have R1 routes in its
  routing table. Naturally, being a helpful route,
  R3 will send R2 the R1 routes (with an additional
  hop).
• I hope we can all see why this would be a Bad
  Thing .

60
RIP - Count to Infinity

• Fortunately, this does not continue forever.
• As far as RIP is concerned, a hop count of 16
  is unreachable. As the route loops, when it
  reaches 16 hop, the route is discarded.

61
RIP - Split Horizon

• To try and stop this type of looping in less than
  (16 30 seconds), Split Horizon was invented.
• In Split horizon, a route is not advertised out
  the interface from which it was learned.

62
RIP - Split Horizon

• In our example, R3 would never have advertised R1
  routes to R3. A routing loop would never have
  been created.

Router 1
Router 3
Router 2
63
RIP - Poison Reverse

• Another anti-loop mechanism is called Poison
  Reverse. In poison reverse, routes are
  advertised with a hop count of 16 out the
  interface from which they were learned.

Router 1
Router 3
Router 2
64
RIP - Poison Reverse

• Split Horizon is preferred over Poison Reverse
  because PR requires additional bandwidth and CPU.

Router 1
Router 3
Router 2
65
RIP Version 1 vs. Version 2

• The major difference between RIPv1 and VIPv2 is
  Classless Inter-Domain Routing (CIDR).

66
RIP Version 1 vs. Version 2

• RIPv2 allows one to advertise inconsistent subnet
  masks, supernets, and discontiguous subnets.
• RIPv2 was described once described by Justin
  Newton as an unsuccessful lobotomy on a
  brain-dead protocol.

67
RIP Version 1 vs. Version 2

• Because RIPv2 still has an inflexible metric,
  wastes bandwidth on route advertisements, has
  slow convergence, can create routing loops, etc.,
  it is still a rarely used protocol.

68
RIP

• Questions

69
OSPF
70
OSPF

• OSPF is the Open Shortest Path First protocol.
• That means it is an Open version of the
  Shortest Path First algorithm, it does not mean
  it tries to open the shortest path first.

71
OSPF

• OSPF version 2 is defined in RFC 2178.
• (Version 1 is no longer in use.)

72
OSPF

• OSPF is a classless (uses CIDR), update-based,
  link-state, open routing protocol.
• These attributes make OSPF the most commonly used
  IGP in use today.

73
OSPF

• OSPF sends its entire routing table upon startup,
  then sends a keep alive every 10 seconds. If a
  topology change occurs, only the changes are
  sent.
• This is much less bandwidth intensive and much
  faster than RIP.

74
OSPF - Link State Protocol

• OSPF is a link state protocol.
• This means that each router keeps a database of
  all the links in its area, and calculates the
  shortest path to each destination network from
  that database.

75
OSPF - Flooding

• In each area, every time a link changes state,
  every router is flooded with Link State
  Advertisements (LSAs) describing the change.
• Every router must run the Dijkstra algorithm to
  re-calculate every route in the area.

76
OSPF - LSAs

• Because this is a beginners class, we will not
  discuss the different types of LSAs.
• A good tutorial for more in-depth OSPF design and
  configuration is
• http//www.cisco.com/warp/public/104/1.html

77
OSPF - Convergence

• This may seem CPU and memory intensive, and it
  does take a great deal more memory and CPU and
  RIP, but it also allows OSPF to converge routes
  in seconds even over large and complex networks.

78
OSPF - Dijkstra Algorithm

• In a nutshell, the Dijkstra Algorithm has each
  router imagine itself as the root of a tree, and
  calculates each successive link as a branch in
  the tree.

79
OSPF - Link Cost

• Each link is assigned a cost. By default this
  cost is 100,000,000 / (speed of link in bps).
  So, the default cost for a FDDI link is 1,
  10BaseT is 10, and a T1 is 64.
• Unfortunately, this does not scale well with
  todays technology.

80
OSPF - Link Cost

• Fortunately, the cost of a link can be set
  manually. In cisco, this is done under each
  interface
• interface serial 0
• ip ospf cost 10

81
OSPF - Dijkstra Algorithm

• A total cost is then calculated for each
  destination prefix. Each prefix is installed
  into the routing table with a next hop relating
  to the lowest cost path.

82
OSPF - Convergence

• When a link changes state, the LSA flood and
  recalculation happen in a very short time,
  usually seconds.
• Because a link change is explicitly stated, there
  are very few routing loops (and for very short
  periods) in OSPF.

83
OSPF - Areas

• To help conserve CPU and RAM, and to limit LSA
  floods, areas were introduced.
• Each router need only know about the links in its
  area, and the link back to Area 0 (zero).

84
OSPF - Areas

• Areas are defined as a 32-bit number, either
  straight decimal (e.g. 123456) or as a dotted
  decimal (e.g. 10.0.0.1).

85
OSPF - Area 0

• Every area must be directly connected to Area 0.

Area 0
Area 1
Area 3
Area 2
86
OSPF - Virtual Links

• A tunnel, or virtual link, can be used when
  direct physical connectivity cannot be achieved.
• This is not the preferred method.

87
OSPF - ABRs

• Routers with an interface in Area 0 and an
  interface in a non-zero area are called Area
  Border Routers or ABRs.
• ABRs aggregate the prefixes for a non-zero area
  and inject the aggregated prefixes into Area 0.

88
OSPF - Area 0

• The routers in Area 0 contain the aggregated
  prefixes for every area.
• Area 0 is sometimes called the Backbone area
  because all inter-area traffic must traverse Area
  0.

89
OSPF - Area 0

• Area 0 places an extreme burden on a network
  designer using OSPF.
• Many networks grow in non-elegant ways, making
  a truly hierarchical network difficult or even
  impossible.

90
OSPF - Multicast

• OSPF uses multicast (224.0.0.x) to propagate its
  routing updates, not broadcast. This reduces the
  CPU requirement on other hosts on the LAN as they
  do not have to process the multicast packet if
  they are not part of the multicast group.

91
OSPF - Neighbors

• OSPF uses neighbor relationships to send routing
  updates.
• If a neighbor relationship cannot be achieved, no
  routing updates will pass.

92
OSPF - DR

• On broadcast media (e.g. Ethernet), OSPF elects a
  Designated Router (DR) and a Backup Designated
  Router (BDR).

93
OSPF - DR

• When updates are sent, each router on the LAN
  sends the updates to the DR (and the BDR), which
  sends one copy to each router.
• This is much better than each router sending a
  copy of each update to each other router.

94
OSPF - BDR

• If the DR is disabled or otherwise does not
  respond to queries, the BDR takes over.

95
OSPF

• Questions

96
EIGRP
97
EIGRP

• EIGRP is the Enhanced Interior Gateway Routing
  Protocol.
• Is was based upon the Interior Gateway Routing
  Protocol (IGRP).

98
EIGRP

• Both EIGRP and IGRP are cisco proprietary
  protocols and do not run on any other router than
  cisco.
• EIGRP propagates route information for IP, IPX
  and AppleTalk, but we will only discuss IP here.

99
EIGRP - DUAL

• EIGRP uses an algorithm called the Distributed
  Update ALgorithm (DUAL).
• You can find out more about EIGRP at
• http//www.cisco.com/warp/public/103/1.html

100
EIGRP - Metric

• EIGRP is essentially a distance-vector protocol.
  The vector is a calculation of four variables,
  each with a static multiplier.
• Metric aB bL cR dD

101
EIGRP - Metric

• These variables are
• Bandwidth (B)
• Load (L)
• Reliability (R)
• Delay (D)

102
EIGRP - Metric

• Recalling our calculation
• Metric aB bL cR dD
• By default, the variables b, c and d are set to
  zero. This leaves the bandwidth as the deciding
  factor in all route computations.

103
EIGRP vs. OSPF

• Because bandwidth is an interface command, EIGRP
  looks a lot like OSPF on this level.
• However, EIGRP has major differences from OSPF.
  For instance, EIGRP and OSPF is a link-state
  protocol while EIGRP is DV.

104
EIGRP vs. OSPF

• Of course, there is also the fact that EIGRP is
  cisco proprietary. This means it cannot be used
  with other vendors routers, such as Lucent, Bay
  Networks, Novell, Microsoft, 3Com, etc.
• For this section, we will assume all cisco gear.

105
EIGRP vs. OSPF

• The largest operational difference between
  EIGRP and OSPF is that EIGRP has no concept of
  Area 0 or a Backbone Area.
• This makes EIGRP much more forgiving of
  evolving networks.

106
EIGRP

• But there are many similarities between EIGRP and
  OSPF. For instance
• EIGRP uses multicast to communicate.
• EIGRP sends only topology changes
• EIGRP keep alives are timed at 10 seconds
• EIGRP converges in seconds, even for complicated
  networks.

107
EIGRP

• Questions

108
iBGP
109
What is BGP?

• BGP is the Border Gateway Protocol, as defined in
  RFC 1771.
• BGP version 4 is a distance vector, classless IP
  routing protocol running over TCP port 179.

110
What is BGP?

• BGP was designed as an Exterior Gateway Protocol
  (EGP). BGP is used to propagate extremely large
  numbers of routes between multiple autonomous
  systems (ASes).
• Most Interior Gateway Protocols (IGPs) have
  faster convergence and better metrics than BGP,
  but are not nearly as scalable.

111
iBGP vs. eBGP

• There are really two kinds of BGP available
• Internal BGP - BGP between peers within the same
  AS.
• External BGP - BGP between peers of different
  ASes.

112
eBGP

• By Default, eBGP peers
• Communicate over directly connected interfaces
• Trade all best routes in the BGP table
• Transmit Prefix, Mask, MED, Origin Code, Next-Hop
  and AS-Path attributes
• Add their ASN to the AS-Path upon transmission

113
iBGP

• By Default, iBGP peers
• Do not need to be directly connected
• Transmit Prefix, Mask, Local-Preference, AS-Path,
  Next-Hop, MED and Origin Code attributes
• Do not modify the AS-Path attribute

114
iBGP

• By Default
• iBGP peers will only propagate routes originated
  by that router or eBGP routes which are best to
  other iBGP peers
• This last point is extremely important to BGP.

115
iBGP

• Assume AS1234 sends route 10.0.0.0/8 to AS5678.
  Router A will send that route to Routers B and C.

B
AS5678
AS1234
A
C
116
iBGP

• When Router B receives 10.0.0.0/8, it will not
  propagate that route to Router C because it was
  learned from an iBGP neighbor. Router C will
  behave similarly.

B
AS5678
AS1234
A
C
117
iBGP

• Furthermore, the Next Hop for 10.0.0.0/8 will be
  the serial interface on the AS1239 router, even
  in Router Bs and Router Cs forwarding table.

B
AS5678
AS1234
A
C
118
iBGP - Next Hop

• Because the Next Hop attribute is not usually a
  directly connected interface, iBGP works
  recursively.
• After the Next Hop is found, a second forwarding
  table lookup is made using the BGP Next Hop as
  the destination.

119
iBGP - Next Hop

• Unfortunately, in ciscos implementation, the
  next hop cannot come from BGP. Therefore, it is
  difficult to use iBGP as the only IGP.
• However, use of iBGP in conjunction with another
  IGP is common and frequently necessary.

120
iBGP - Multiple IGPs

• Most networks use iBGP and a second IGP to
  control routing.
• For instance combining OSPF and iBGP, the next
  hop information for every BGP route can be
  learned in each router.

121
iBGP

• Questions

122
Interaction with BGP
123
Multiple Routers and iBGP

• Things get significantly more complex when
  multiple BGP speaking routers are involved.
• ISP1 loopback 10.0.0.1
• ISP2 loopback 10.0.0.2
• ISP3 loopback 10.0.0.3

ISP1
Upstream A (2828)
ISP3
ISP2
Upstream B (701)
124
Interaction with BGP

• This section deals with the Interaction between
  IGPs and BGP.
• Only multi-homed networks should be dealing with
  BGP.

125
Interaction with BGP

• If a network has only one exit router, there is
  really very little interaction between BGP and
  IGPs.
• A default route in each router is usually
  sufficient.

126
Interaction with BGP

• For instance, if R1 is the only exit router, R2
  and R3 need no BGP information. Even R1 has no
  interaction between the IGP and BGP.

Router 2
Upstream A
Router 1
Router 3
Upstream B
127
Multiple Routers and iBGP

• Things get significantly more complex when
  multiple BGP speaking routers are involved.
• R1 loopback 10.0.0.1
• R2 loopback 10.0.0.2
• R3 loopback 10.0.0.3

Router 1
Upstream A
Router 3
Router 2
Upstream B
128
Multiple Routers and iBGP

• R1 and R2 must speak iBGP to get optimal
  routing.
• What may not be immediately obvious is that R3
  must also speak iBGP.

Router 1
Upstream A
Router 3
Router 2
Upstream B
129
Multiple Routers and iBGP

• What might happen if R3 does not speak BGP, but
  R1 and R2 do speak iBGP?
• Assume R3 does not speak iBGP with R1 and R2.
  When a packet comes into R3, there is no way for
  R3 to know which border router to send the packet.

130
Multiple Routers and iBGP

• Assume R3s route table gives R1 as the next hop
  for this packet for some reason. (Load
  balancing, default route, black magic, .) So R3
  sends the packet to R1.

R1
Upstream A
R3
R2
Upstream B
131
Multiple Routers and iBGP

• Assume further that the actual best route for
  that destination is through Upstream B. Well,
  since R1 gets to Upstream B through R3, the
  packet gets sent back.

R1
Upstream A
R3
R2
Upstream B
132
Routing Loop

• Unfortunately, as far as R3 is concerned, this is
  a new packet, so R3 will send the packet the same
  place it did last time - R1. This will cause a
  routing loop.

ISP1
Upstream A (2828)
ISP3
ISP2
Upstream B (701)
133
Multiple Routers and iBGP

• To stop this from happening, run iBGP on R3, and
  R3 will choose the correct exit router and send
  the packet accordingly.
• This will also help on your internal bandwidth
  costs, even if the exit routers have direct
  connectivity.

134
Configuration Sample R1

• router BGP 15000
• no synchronization
• neighbor 172.16.0.1 remote-as 2828
• neighbor 172.16.0.1 next-hop-self
• neighbor 172.16.0.1 soft-reconfig in
• neighbor 172.16.0.1 filter-list 1 out
• neighbor 10.0.0.2 remote-as 15000
• neighbor 10.0.0.2 update-source loopback0
• neighbor 10.0.0.3 remote-as 15000
• neighbor 10.0.0.3 update-source loopback0

135
Configuration Sample R2

• router BGP 15000
• no synchronization
• neighbor 172.16.1.1 remote-as 701
• neighbor 172.16.1.1 next-hop-self
• neighbor 172.16.1.1 soft-reconfig in
• neighbor 172.16.1.1 filter-list 1 out
• neighbor 10.0.0.1 remote-as 15000
• neighbor 10.0.0.1 update-source loopback0
• neighbor 10.0.0.3 remote-as 15000
• neighbor 10.0.0.3 update-source loopback0

136
Configuration Sample R3

• router BGP 15000
• no synchronization
• neighbor 10.0.0.2 remote-as 15000
• neighbor 10.0.0.3 update-source loopback0
• neighbor 10.0.0.3 remote-as 15000
• neighbor 10.0.0.3 update-source loopback0

137
Update Source Loopback0

• You may have noticed that the update source for
  the iBGP peers is loopback0.
• This is done so that no one interface can
  interrupt the flow of BGP routes.

138
Update Source Loopback0

• If you tie the iBGP neighbor relationship to the
  interface between the two routers, and that
  interface goes down, the iBGP session will die
  even if there is an alternate path.
• Remember, iBGP peers do not need to be directly
  connected.

139
IGP and BGP

• However, BGP has no explicit knowledge of the
  internal routing. BGP cannot calculate the next
  hop if it is not a directly connected interface.
  BGP cannot even figure out where the other
  loopback interfaces are.
• This is where the IGP comes into play.

140
IGP and BGP

• So, in a multiple exit network, you actually need
  to run three protocols - eBGP, iBGP and another
  IGP.

141
No Synchronization

• The command no synchronization is necessary in
  multiple router configurations.
• Without this command, a router will not advertise
  a route to an external peer unless the route is
  local or exists within the IGP.

142
Load Balancing with BGP

• Another important interaction between IGPs and
  BGP is for certain types of BGP load balancing.

143
Load Balancing with BGP

• There are two types of load balancing with BGP.
  The most common type is to shape the traffic
  from each of two upstreams to create optimal
  routing. We have already covered that type of
  load balancing in the previous example.

144
Load Balancing with BGP

• The other load balancing with BGP is when there
  are two lines to the same upstream.
• Because BGP by default picks on Next Hop for each
  prefix, in this situation one line would be used
  and the other would be empty. (Unless the first
  line became unusable.)

145
Load Balancing with BGP

• It is possible to load balance over two lines to
  the same upstream with BGP.
• This can be done either through eBGP-Multihop or
  BGP Multi-Path support. We will discuss the
  eBGP-Multihop method in this class.

146
Load Balancing with BGP

• Load balancing with eBGP-Multihop is not actually
  using BGP to load balance, it is using IGP to
  load balance.
• The simple explanation is you create a BGP Next
  Hop which has multiple IGP routes and use the IGP
  to load balance across those routes.

147
Load Balancing with BGP

• Using a simple test case, an ISP has two T1s to
  their upstream. Running normal BGP between the
  upstream, all and the ISP would default to the T1
  with the lowest IP address.

T1
ISP
Upstream
T1
148
Load Balancing with BGP

• By using eBGP-Multihop to the looback interfaces
  and next-hop-self, a recursive lookup in the IGP
  is needed.

ISP
Upstream
l0
l0
149
Load Balancing with BGP

• Think of BGP as a two step process.
• First, you find the BGP Next Hop.
• Second you find the route to the BGP Next Hop.
• The packet it then routed to the BGP Next Hop
  where it will be routed to the final destination.

150
Interaction with BGP

• Questions

151
Examples of Networks Using Each Protocol
152
Network Examples

• In this section we will be building a simple
  network with each protocol.
• Each network has a protocol which is best for
  that network. Hopefully your network will fall
  into one of these categories.

153
Network Examples - RIP

• In the following network, RIP is the most useful
  protocol.
• Here we have a hub site with multiple stub
  sites.

154
Network Examples - RIP

• RIP has a useful property that none of the other
  protocols have - no neighbor relationship.
• This means a router can hear, but does not have
  to speak.

155
Network Examples - RIP

• In this case, each of the stub sites can
  broadcast their network information, and point
  default at the hub site.

156
Network Examples - RIP

• The hub site need only listen to the
  announcements, not send back any information.

157
Network Examples - RIP

• It turns out that for one or two prefixes, a
  single RIP advertisement every 30 seconds
  actually takes less bandwidth than three OSPF
  keep alives.
• Plus, all the stub sites have virtually no CPU or
  RAM requirements.

158
Network Examples - OSPF

• We have already seen the OSPF perfect network.
• It is a network that is hierarchically laid out,
  with routers logically grouped.

159
Network Examples - OSPF

• OSPF was essentially designed for something that
  looks like a FDDI backbone with a bunch of stub
  ethernet networks hanging off it.
• Fortunately, it works for more cases than that.

160
Network Examples - OSPF

• Assume each cloud is a collection of routers in
  a semi- or full-mesh.

Area 0
Area 1
Area 3
Area 2
161
Network Examples - OSPF

• If the routers are grouped and there is a
  backbone area, LSA flooding is optimized.

Area 0
Area 1
Area 3
Area 2
162
Network Examples - EIGRP

• A network with multiple areas, but no single
  backbone area is one example of a network good
  for EIGRP.
• Because the network is complicated, RIP and
  static routes are unacceptable.

163
Network Examples - EIGRP

• However, OSPF does not work well in such an
  environment. There is no backbone area.

164
Network Examples - EIGRP

• Of course, work arounds are possible,
• However, EIGRP can handle the network elegantly
  without work arounds.

165
Network Examples - EIGRP

• It turns out that EIGRP is actually optimal for
  most real life networks - except that it is
  cisco proprietary.
• EIGRP is optimal for so many networks because it
  has no stringent design criteria like OSPFs Area
  0.

166
Network Examples

• Questions

167
Examples of Networks Using Multiple Protocol
168
Examples - Multiple Protocols

• We have already covered the case of one example
  where multiple IGPs are optimal - the use of iBGP
  with a second IGP.

169
Examples - Multiple Protocols

• A less common case is where one is forced to use
  an IGP at a location that would not be optimal
  for the entire network.
• For instance, if one POP has terminal servers
  which only support RIP, but the network is
  relatively complex.

170
Examples - Multiple Protocols

• If we take our previous OSPF example and add a
  LAN with RIP

LAN with RIP
171
Examples - Multiple Protocols

• In this example, the R1 runs both RIP and OSPF,
  redistributing the RIP into OSPF.

LAN with RIP
R1
172
Examples - Multiple Protocols

• The other routers only speak OSPF and do not need
  to deal with the secondary protocol.

LAN with RIP
R1
173
Examples - Multiple Protocols

• There are several pitfalls, and one must be
  careful redistributing between two protocols, but
  it is far better than using WAN bandwidth for RIP
  and running multiple protocols on all routers to
  accommodate a single legacy device.

174
Multiple Protocols

• Questions

175
Questions