1
|
- Ch. 11 – Routing Theory – Part 2
- CCNA Semester 2
- Rick Graziani, Instructor
- Feb. 5, 2002
|
2
|
- Instructors:
- Most of the information in the presentation is a rewrite of the curr=
ent
Semester 2, Ch. 11 Routing, including:
- Clarification of all topics
- Additional information on most topics
- Many new topics
- If you have any questions or comments, please email:
- Rick Graziani
- Cabrillo College
- graziani@cabrillo.cc.ca.us
|
3
|
- Much of the information in this presentation be reinforced with more
detail and more examples when we discuss the additional presentations:
- Ch.12 Routing Protocols
- Additional Semester 2 Presentations
- The Routing Table Structure
- Discard Routes
- Static Routing – Additional Information
- And again in the CCNP Semester 5 Advanced Routing class
- Understanding the behavior and affect of routing protocols is the
difference between people who are “paper CCNAs” and those
people who have the skills and knowledge that the CCNA exam is suppo=
se
to represent.
- This presentation, like the others, is designed to help give you tho=
se
knowledge and skills.
|
4
|
- Part I. Routing Basics=
and
Static Routing
- Basic Concepts:
- Network Layer
- IP Routing Table
- Path Determination
- Routed Protocols versus Routing Protocols
- Network Layer Protocol Operations
- Path Switching (Introduction)
- Multiprotocol Routing
- IP Routing Table and Directly Connected Networks
- Static Routing
- Configuring Static Routes
- Static Routing in the Real-world
- Default Static Routes
- Recursive Lookups
- Static Routes and the Routing Table Process
- Advantages and Disadvantages of Static Routing
|
5
|
- Part II. Routing Theor=
y and
Dynamic Routing Operations
- Dynamic Routing Operations
- Routing Metrics
- Classes of Routing Protocols
- Convergence
- Distance Vector Routing Protocols
- Distance Vector Concepts
- Distance Vector Network Discovery
- Simple Split Horizon (Introduction)
- Distance Vector Network Discovery with Split Horizon
- Network Discovery FAQs
- Triggered Updates
- Routing Loops
- Count to Infinity
- Defining a Maximum
- Split Horizon
- Split Horizon with Poison Reverse
- Holddown Timers
- TTL – IP’s Time-To-Live Field
|
6
|
- Part III. Routing Theo=
ry and
Dynamic Routing Operations (continued)
- Link-State Routing Protocols
- Link-state Concepts
- Link-state Routing Protocol History
- Theory of Link-State Routing Protocols
- Mathematical Point Of View
- Link-state Concepts
- 1. Flooding of Link-State Information
- 2. Building a Topological Database
- 3. Shortest-Path-First (Dijkstra’s) Algorithm
- 4. Shortest-Path-First Tree
- 5. Routing Table
- Exercise: From Link-State Flooding to Routing Tables
- Hello Messages and LSAs (Link-State Advertisements)
- Topology Changes
- Link-State Concerns
- Problem: Link-State Updates – LSA Sequence Numbers
- Comparing Distance Vector and Link State Routing Protocols
- For Additional Information on Link State Routing
|
7
|
- Part III. Routing Theo=
ry and
Dynamic Routing Operations (continued)
- Hybrid Routing Protocols
- Concepts
- EIGRP (not IS-IS)
- Path Switching
- Example: Host X to Host Y (with three routers in between)
- LAN-to-LAN Routing
- LAN-to-WAN Routing
- Cisco Router Configuration
- Summary
- Topics (Review)
|
8
|
- Part II. Routing Theor=
y and
Dynamic Routing Operations
- Dynamic Routing Operations
- Routing Metrics
- Classes of Routing Protocols
- Convergence
- Distance Vector Routing Protocols
- Distance Vector Concepts
- Distance Vector Network Discovery
- Simple Split Horizon (Introduction)
- Distance Vector Network Discovery with Split Horizon
- Network Discovery FAQs
- Triggered Updates
- Routing Loops
- Count to Infinity
- Defining a Maximum
- Split Horizon
- Split Horizon with Poison Reverse
- Holddown Timers
- TTL – IP’s Time-To-Live Field
|
9
|
|
10
|
- The success of dynamic routing depends on two basic router functions=
:
- 1. maintenance of a ro=
uting
table
- 2. timely distribution=
of
knowledge, in the form of routing updates, to other routers
|
11
|
- Dynamic routing relies on a routing protocol to share knowledge among
routers.
- A routing protocol defines the set of rules used by a router when it
communicates with neighboring routers. For example, a routing protoc=
ol
describes:
- how to send updates
- what knowledge is contained in these updates
- when to send this knowledge
- how to locate recipients of the updates
|
12
|
- When a routing algorithm updates a routing table, its primary object=
ive
is to determine the best information to include in the table.
- Each routing algorithm interprets what is best in its own way. The
algorithm generates a number, called the metric value, for each path
through the network.
- Typically, the smaller the metric number, the better the path.
|
13
|
- These metrics will be discussed later or in CCNP Advanced Routing, w=
ith
their appropriate Routing Protocols:
- RIP – hop count
- IGRP – bandwidth, delay, reliability, load
- EIGRP – bandwidth, delay, reliability, load
- OSPF – bandwidth
- BGP – attribute values and shortest path
|
14
|
- Most routing algorithms can be classified as one of two basic
algorithms:
- distance vector
- link state.
- The distance-vector routing approach determines the direction (vecto=
r)
and the cost or metric (distance) to any link in the internetwork.=
li>
- RIP, IPX RIP and IGRP (CCNA)
- AppleTalk, RTMP and others (non-CCNA)
|
15
|
- The link-state (also called shortest path first) approach re-creates=
the
exact topology of the entire internetwork (or at least the portion in
which the router is situated).
- OSPF
- IS-IS
- Note: Current CCNA material discusses only distance-vector protocols,
however we will introduce OSPF later in the semester. CCNP Advanced Routing exami=
nes
link-state routing and OSPF in much more detail.
|
16
|
- The balanced hybrid approach combines aspects of the link-state and
distance-vector algorithms.
- These are really distance-vector routing protocols which apply some =
of
the advantages of a link-state routing protocols, and also known as =
advanced-distance-vector
routing protocols.
- EIGRP
- Note: Many of the concepts we will learn about IGRP (beyond the norm=
al
on-line curriculum) apply to EIGRP.=
CCNP Advanced Routing discusses EIGRP.
|
17
|
- When all routers in an internetwork are operating with the same
knowledge, the internetwork is said to have converged.
- Fast convergence is a desirable network feature because it reduces t=
he
period of time in which routers would continue to make
incorrect/wasteful routing decisions.
|
18
|
|
19
|
- Part II. Routing Theor=
y and
Dynamic Routing Operations
- Dynamic Routing Operations
- Routing Metrics
- Classes of Routing Protocols
- Convergence
- Distance Vector Routing Protocols
- Distance Vector Concepts
- Distance Vector Network Discovery
- Simple Split Horizon (Introduction)
- Distance Vector Network Discovery with Split Horizon
- Network Discovery FAQs
- Triggered Updates
- Routing Loops
- Count to Infinity
- Defining a Maximum
- Split Horizon
- Split Horizon with Poison Reverse
- Holddown Timers
- TTL – IP’s Time-To-Live Field
|
20
|
- Distance Vector Concepts
- The mathematical basis of the distance-vector routing protocols is t=
he Bellman-Ford
algorithm.
- Pure distance-vector routing protocols suffer from long convergence
times and possible temporary routing loops (more in a few moments).<=
/li>
- There are remedies to some situations that may cause these problems
which we will examine in a moment.
|
21
|
- Network Discovery and Routing Table Maintenance
- Distance-vector-based routing algorithms pass periodic copies of a
routing table between adjacent routers, from router to router. (RIP every 30 seconds, IPX =
RIP
every 60 seconds, IGRP every 90 seconds).
- These regular updates between routers help routers discover each
other’s networks and communicate topology changes.
- Routers only learn about other networks from adjacent routers, their
directly connected neighbors.
- Router D learned about Router A’s network 172.16.0.0/16 from
Router C, who learned it from Router B, who learned it from Router A=
.
- This is why distance-vector routing protocols are also known as rout=
ing
by rumor.
|
22
|
- Distance-vector routing protocols do not allow routers to know the
topology of the network, as they only know how far a network is
(distance: hops) and which way to forward the packet (vector: exit
interface). (Link-state
routing protocols allow routers to see the exact network topology
– later.)
- “The algorithm eventually accumulates network distances so tha=
t it
can maintain a database of network topology information.” (On-=
line
curriculum)
- Distances (hops) are cumulative from one router to the next, however
RIP does not keep a database of network topology information (except
for demand circuits in IOS 12.0).
|
23
|
- How do routers learn about other networks and determine the best rou=
tes
to these networks?
- We will now look at the concepts of how this happens.
- In Chapter 12 Routing Protocols, we will look at how RIP does this a=
nd
view it happening!
|
24
|
- Initial Routing Tables
- The first routes entered in the routing table are directly connected
networks, assigning a metric or cost of “0” (hop count for RIP).
- See previous slides on IP Routing Table and Directly Connected
Networks.
- 00:28:56: RT: add 192.168.2.0/24 via 0.0.0.0, connected metric [0/0=
]
- 00:28:56: RT: interface Ethernet0 added to routing table
- Remember, these interfaces must be “up” and “up=
221;
- The next step is for routers to share their complete routing tables =
with
any and all directly connected neighboring routers.
- Distance-vector routing protocols do not maintain formal relationshi=
ps
with neighboring routers, I.e. they do not know who their neighboring
routers are.
- So how do they know who to send their routing tables to? Distance-vector routing pro=
tocols
use a broadcast or multicast address to send out routing updates,
although you can specify a unicast address. (later)
|
25
|
- Sending Routing Tables to Neighbors
- (Later, we will see another example with real IP addresses.)
- In their routing updates to the neighboring router(s), distance-vect=
or
routing protocols include the following information for each network=
in
their routing table:
- Network address – This would normally be an ip network addres=
s.
- The metric or cost (with RIP this is the number of hops, but can be
other metrics for other distance-vector routing protocols).
- Since we are using hops in this example, RIP increments the hop co=
unt
by one in its routing table before sending out the routing update.=
- Next-hop address – This would normally be the ip address of t=
he
interface from which the routing update was sent.
|
26
|
- Routers Update their Routing Tables
- A router will enter this update into its routing table if:
- It is a new route, a network which is not currently in its routing
table.
- It is an existing route, a network which is currently in its routing
table, but this update has a better (smaller) metric (fewer hops).<=
/li>
- Note: If the update
contains a route to an existing route, with the same metric (hops),=
but
via a different interface, the router may or may not add it to its
routing table depending upon whether or not the routing protocol is
providing load balancing (later).&=
nbsp;
RIP does provide this.
- A router will not enter this update into its routing table if:
- It is an existing route with a worse metric.
|
27
|
- Routers Update their Routing Tables (continued)
- RTA:
- Ignores the route to X because it has an existing, better route with=
a
cost=3D0.
- Accepts the route to Y because it didn’t exist in the routing
table.
- RTB (from RTA):
- Accepts the route to W because it didn’t exist in the routing
table.
- Ignores the route to X because it has an existing, better route with=
a
cost=3D0.
- RTB (from RTC):
- Ignores the route to Y because it has an existing, better route with=
a
cost=3D0.
- Accepts the route to Z because it didn’t exist in the routing
table.
- RTC:
- Accepts the route to X because it didn’t exist in the routing
table.
- Ignores the route to Y because it has an existing, better route with=
a
cost=3D0.
|
28
|
- Next Round of Routing Updates
- We have still not reached convergence, because all of the routers do=
not
have complete and accurate network information.
- Which router does have complete information? Which ones do not?
- In the next round, routers must forward their new routing tables in =
the
form of routing updates, to their directly connected neighbors.
- Remember, with the distance-vector routing protocol RIP, the router
increments the number of hops in its own routing table by one, before
sending out the routing update.&nbs=
p;
- “If it is one hop for me to get there, and you are
getting there via me, then it is two hops for you.”
|
29
|
- Routers Update their Routing Tables - again
- RTA:
- Ignores the route to X because it has an existing, better route with=
a
cost=3D0.
- Ignores the route to Y because it already has that route, with the s=
ame
cost.
- Ignores the route to W because it has an existing, better route with=
a
cost=3D0.
- “I am not going to send you the packet, so you can send it ba=
ck
to me, …”
- Later, we will see that split-horizon prohibits this route from bei=
ng
sent.
- Accepts the route to Z because it didn’t exist in the routing
table.
|
30
|
- Routers Update their Routing Tables – again (continued)
- RTB:
- Ignores all routes from both RTA and RTC, because it already has tho=
se
routes, with the same costs.
- No new information.
|
31
|
- Routers Update their Routing Tables - again
- RTC:
- Ignores the route to X because it already has that route, with the s=
ame
cost.
- Ignores the route to Y because it has an existing, better route with=
a
cost=3D0.
- Accepts the route to W because it didn’t exist in the routing
table.
- Ignores the route to Z because it has an existing, better route with=
a
cost=3D0.
- “I am not going to send you the packet, so you can send it ba=
ck
to me, …”
|
32
|
- Convergence!
- All of the routers now have a consistent and accurate view of the
network.
- Later, we will see how RIP handles this operation.
|
33
|
- Split Horizon Rule
- Before we continue looking at routing tables and network discovery,
using real ip addresses, let’s take a look at the split horizon
rule.
- “The effect of split horizon is that a router will send out
different routing messages on different interfaces. In effect a router never se=
nds
out information on an interface that it learned from that interface.=
”
(Lewis, Cisco TCP/IP Routing)
- As we will see later in this presentation, split horizon helps preve=
nt
routing loops. (Discus=
sed in
much more detail soon.)
- For now, we will see that split horizon means that the router does n=
ot
send out all of the information in the routing table to its neighbor=
s.
- Note: Usually, split horizon is enabled and can be disabled.
|
34
|
- Network Discovery with Split Horizon
- First of all, notice we are using real ip addresses.
- In the routing updates, next-hop ip addresses for the networks are s=
ent
to the neighboring router specifying the address it can use to forwa=
rd
packets to.
- The split horizon rule also affects common networks between two rout=
ers.
- To the router, a directly connected network is known via its own
interface, so it does not include that network in routing updates se=
nt
out that same interface.
- In other words, the router does not send information about a directly
connected network out the interface of that directly connected netwo=
rk.
|
35
|
- Network Discovery with Split Horizon - First Round of Updates
- “The effect of split horizon is that a router will send out
different routing messages on different interfaces. In effect a router never se=
nds
out information on an interface that it learned from that interface.=
”
(Lewis)
- RTA’s routing update sent out serial 0 to RTB
- Includes the network 10.1.1.0/24 which RTB can reach via 10.1.1.1.=
li>
- Split horizon: Does not include the 10.1.2.0/24 network because that
network was learned via serial 0 (interface serial 0, ip address
10.1.1.1 …) – a common network between RTA and RTB.
|
36
|
- Network Discovery with Split Horizon - First Round of Updates
- RTB’s routing update sent out serial 0 to RTA
- Includes the network 10.1.3.0/24 which RTA can reach via 10.1.2.2.=
li>
- Split horizon: Does not include the 10.1.2.0/24 network. Split horizon blocks the
10.1.2.0/24 update from being sent to RTA with a hop count of
“1.” (Note=
: To
keep the diagrams less cluttered, omission of the proper red/blue ar=
row
means split horizon is in affect, same as the “X.”)
- RTB’s routing update sent out serial 1 to RTC
- Includes the network 10.1.2.0/24 which RTC can reach via 10.1.3.1.=
li>
- Split horizon: Does not include the 10.1.3.0/24 network.
|
37
|
- Network Discovery with Split Horizon - First Round of Updates
- Same with updates from RTC and RTD.
- Your Turn:
- Write out the new routing tables for each router after this round.=
li>
- Also, find any mistakes I might have made J=
font>
|
38
|
- Answer – Do your routing tables look like these?
- Now – What do the next round of routing updates look like? Show the routes which are s=
ent
(propagated) and those that are not sent because of split horizon.=
li>
|
39
|
- Answer – Do your routing updates look like these?
- Again, omission of the red/blue arrow means split horizon is in affe=
ct.
- For example, RTB is not sending the route 10.1.1.0/24 to RTA to tell=
RTA
it can get to 10.1.1./24 in 2 hops via RTB. - This would make sense!
- Split horizon - router never sends out information on an interface t=
hat
it learned from that interface
- Now – What do the routing tables look like?
|
40
|
- Answer – Do your routing tables look like these? Convergence?
- Note: Newest routing table entries are at the bottom of the routing
tables in these diagrams.
- Now – What do the next round of routing updates look like and =
the
routing tables? (We=
217;ll
finish this up J )<=
/font>
|
41
|
- Answer – Do your routing tables look like these?
- Convergence? – YES!
|
42
|
|
43
|
- FAQs – Network Discovery
- Q: How often does initial network discovery happen?
- A: Only when the network comes first comes up.
- Q: Do routers share routing table information after network discover=
y?
- A: Yes, distance-vector routing protocols share their entire routing
tables periodically (with or without split horizon enabled). Distance vector routing pro=
tocols
on Cisco routers by default use split horizon with poison reverse
(discussed in the next section).&nb=
sp;
Depending upon the distance-vector routing protocol, the
frequency of the updates will happen for RIP every 30 seconds, IPX R=
IP
every 60 seconds, and IGRP every 90 seconds.
- Q: What happens when there is a change in the topology, link goes do=
wn,
new network is added, new router, is added, etc.?
- A: Let’s take a look.
|
44
|
- Triggered Updates
- Routers do not have to wait for the periodic update to hear about
changes in the network topology.
- Improvements to the distance-vector algorithm is typically made in
distance-vector routing protocols, like RIP, to include triggered
updates.
- Even with triggered updates, large distance vector networks can suff=
er
from long convergence times in some situations.
|
45
|
- Triggered updates: (continued)
- Triggered updates are sent whenever a router sees a topology change =
or a
change in routing information (from another router).
- The router does not have to wait for the period timer, but can send =
them
immediately.
- Triggered updates do not need to include the entire routing table but
only the modified route(s).
|
46
|
- Triggered updates:
(continued)
- Triggered updates must still be sent to adjacent routers, from route=
r to
router, like other routing updates.
- Most distance-vector routing protocols limit the frequency of trigge=
red
updates so that a flapping link does not put an unnecessary load on =
the
network. (RIP: random 1 to 5 seconds)
- Typically, triggered updates can be “triggered” by:
- Interface transition to the up or down state
- A route has entered or exited an unreachable (down) state (later)=
li>
- A new route is installed in the routing table
|
47
|
- Routing Loops
- Distance vector routing protocols are simple in their implementaton =
and
configuration, but this comes at a price.
- Pure distance vector routing protocols suffer from possible routing
loops.
- Routing loops can cause major network problems, from packets getting
lost (blackholed) in your network, to bringing down your entire netw=
ork.
- Several remedies to have been added to distance-vector algorithms to
help prevent routing loops including:
- Split horizon
- Hold-down timers
- Defining a maximum metric
|
48
|
- Routing Loops (continued)
- What can cause routing loops?
- Routing loops can occur when there are:
- Incorrect or inconsistent routing updates due to slow convergence a=
fter
a topology change. (E=
xample
coming up next.)
- Incorrect or incomplete routing information (see presentation on
Discard Routes)
- Static routes incorrectly configured with an intermediate address w=
hich
does not become resolved in the routing table. (see presentation on
Static Routes – Additional Information)
|
49
|
- Routing Loop Example
- Assume for the remainder of this example that Router C’s prefe=
rred
path to network 1 is by way of Router B.
- Router C’s routing table has a distance of 3 to network 1 via
Router B.
|
50
|
- Network 1 Fails
- Router E sends an update to Router A.
- Router A stops routing packets to network 1.
- But Routers B, C, and D continue to do so because they have not yet =
been
informed about the failure.
- Router A sends out its update.
- Routers B and D stop routing to network1, (via Router A).
- However, Router C is still not updated.
- To router C, network 1 is still reachable via router B.
|
51
|
- Router C sends a periodic update to Router D
- Router C sends a periodic update to Router D indicating a path to
network 1 (by way) of via Router B. (4 hops).
- Router D’s Routing Table information for Network 1
- Current path to Network 1 =3D Unreachable (down)
- Information from Router C: =
span>Network
1 : 4 hops by way of Router C
- Normally, RouterD ignores this routing information because it usually
has a better route, 2 hops, via Router A, but this route is now down=
.
- Router D changes its routing table to reflect this (good) better, but
incorrect information, Network 1 by way of Router C (4 hops)
- Router D propagates the information to Router A.
|
52
|
- Routers A changes its routing table
- Router A adds new route to its routing table, get to Network 1 by wa=
y of
Router D (5 hops).
- Propagates the information to Routers B and E.
- Router B (and Router E) change their routing tables
- Router B now believes it can get to Network 1 by way of Router A (6
hops).
- Wow! I was about to te=
ll
Router C that Network 1 was down via Router B, but now I have new
information!
- Propagates the incorrect information to Router C.
|
53
|
- Router C changes its routing table
- Router C still believes it can get to Network 1 by way of Router B (7
hops).
- Of course now it believes it is 7 hops instead of 3.
- Propagates the newer but still incorrect information to Router D.
- Here we go again!
- Data packets destined for Network 1 get caught in a routing loop, fr=
om
Routers A to D to C to B to A to D etc.
- As routing updates continue between the routers, the hop count gets
greater – to infinity?
(Not quite – we will see in a moment.)
|
54
|
- Counting to Infinity
- The routing loop we just saw creates another problem, known as
“Counting to Infinity.”
- This condition, called count to infinity, loops packets continuously
around the network in spite of the fundamental fact that the destina=
tion
network, Network 1, is down.
- While the routers are counting to infinity, the invalid information
allows a routing loop to exist.
- Without countermeasures to stop the process, the distance vector
(metric) of hop count increments each time the packet passes through
another router. - These packets loop through the network because of
wrong information in the routing tables.
|
55
|
- Solution to Counting to Infinity: Defining a Maximum metric
- Distance vector routing protocols remedy the problem by limiting the
maximum number of hops for any route in the routing table.
- When the distance vector routing protocol has a route with a metric =
that
is more than its maximum-value, it is denoted as “infinityR=
21;
and the route is considered “unreachable.”
- For RIP the maximum-value is 15 (hops), infinity is 16 (hops).
- For IGRP the maximum-value 100 (hops), infinity is 101 (hops).
- IGRP uses bandwidth, delay, reliability and load for its metric in
determining best path.
- IGRP does not use hop count as this metric. Hop count is only used by =
IGRP
to stop the counting to infinity behavior. (more later)
|
56
|
- Solution to Counting to Infinity: Defining a Maximum metric
- Remember that distance vector routing protocols like RIP, increment =
the
metric (hop count) before sending the routing update to their adjace=
nt
routers.
- After incrementing the hop count, if the metric (hops) is less than=
15,
routing updates to other adjacent routes will receive a valid route=
for
this network from this router.
- After incrementing the hop count, if the metric (hop count) is equa=
l to
15, this router will be able to route packets to this network, 15 h=
ops
away, but routing updates to other adjacent routers will have the
incremented hop count of 16 (infinity). - This means other routers=
cannot
reach this network via this router.
- After incrementing the hop count, if the metric (hop count) is equa=
l to
16,“infinity”, this router will not be able to route
packets to this network.
Routing updates to other adjacent routers will also have the=
hop
count of 16 (infinity), which means they cannot reach this network =
via
this router.
- There is another situation where the router itself with modify the h=
op
count to infinity – split horizon with poison reverse. –
Coming up next!
|
57
|
- FAQs – Defining a maximum-value
- Q: Why does RIP use ahop count as the route metric, and why is its
maximum value limited to 15?
- A: When RIP was designed and implemented, dynamic routing protocols =
were
not widely used. Inste=
ad,
networks relied mostly on static routing. RIP, even with its
hop-count-metric – which seems very poor to us today – w=
as
quite a big improvement.
Counting intermediate routes is the simplest method to measure
the quality of routes.
Setting the infinity value for the metric is always a problem=
of
choosing between wider networks and faster convergence when the prot=
ocol
starts counting. When RIP was invented, it seemed unlikely to have a
network with the maximum diameter more more than 15 routers, so 16 w=
as
chosen as the infinity value. (Zinin, Cisco IP Routing)
|
58
|
- Note: The information contained in the on-line curriculum is not
accurate regarding split-horizon.&n=
bsp;
The reason Router D does not send updates regarding Network 1=
to
Router A is because of hold-down timers and not because of split
horizon.
- The following slides along with the previous slides on split horizon
should clarify this concept for you.
|
59
|
- Split Horizon Rule - Reminder
- “The effect of split horizon is that a router will send out
different routing messages on different interfaces. In effect a router never se=
nds
out information on an interface that it learned from that interface.=
”
(Lewis, Cisco TCP/IP Routing)
- Earlier we saw that split horizon meant that the router does not send
out all of the information in the routing table to its neighbors.
- If a router learned about a network from an adjacent router, it does
not include that network in its routing updates to that neighbor.
(See previous slides, Netw=
ork
Discovery and Split Horizon.)
- As we will now see split horizon also helps prevent routing loops.=
li>
|
60
|
- Split Horizon Rule – Avoiding Routing Loops
- Routers RTA and RTB have their initial routing tables and are ready =
to
exchange routing information via a distance-vector routing protocol =
like
RIP.
- Split Horizon disabled
- If split horizon were disabled the routing updates would include all=
of
the networks in their routing tables including their directly connec=
ted
networks and any networks learned from any interface.
|
61
|
- Split Horizon Disabled
- After the initial exchange of updates everything in the routing tabl=
es
look fine.
- Because split horizon disabled, the 10.1.2.0/24 network is sent by b=
oth
routers, but neither router includes the other’s route to
10.1.2.0/24 (1 hop) in the routing table, because it has a current r=
oute
with a better metric of 0.
|
62
|
- Split Horizon Disabled
- After the next exchange of updates everything in the routing tables =
look
fine and the routing tables are converged.
- Because split horizon disabled, the=
besides the 10.1.2.0/24 network, the networks learned from the
other router in the previous update is also sent by both routers.
- However, neither router includes the those networks, because it has a
current route with a better metric of 0.
|
63
|
- Split Horizon Disabled – 10.1.3.0/24 down
- Note: Routing tables are not sent at the exactly same time. We will
learn about this in Ch. 12 Routing Protocols, that this is done on
purpose to avoid collisions on broadcast networks like Ethernet.
- Here, the 10.1.3.0/24 network fails, and before RTB sends out its
routing update, RTB receives a routing update from RTA.
|
64
|
- Split Horizon Disabled – 10.1.3.0/24 down
- RTB notices that it has a route to 10.1.3.0/24 via RTA. Even though it is 2 hops it=
is
certainly better than its current situation of “unreachable=
221;
so it accepts this better, but incorrect information from RTA.
- RTB now forwards all packets destined for 10.1.3.0/24 to RTA at
10.1.2.1.
- RTA receives these packets and forwards them to RTB at 10.1.2.2.
- RTB forwards them back to RTA at 10.1.2.1.
- And so on! The packets=
get
blackholed in this routing loop.
|
65
|
- Split Horizon Disabled – 10.1.3.0/24 down
- Meanwhile, its RTB’s turn to send its routing update.
- RTB increments the hop count to 10.1.3.0/24 to 3 hops and sends it to
RTA.
- When RTA sends out its next routing table it will increment the hop
count to 10.1.3.0/24 to 4 hops and sends it to RTB.
- And on and on, until “infinity” which in RIP is 16 hops.=
|
66
|
- Split Horizon Disabled
- Once both routers have 16 hops for 10.1.3.0/24, they will both mark =
this
network as unreachable and discontinue forwarding, drop, packets to =
this
network.
- This temporary routing loop can be easily avoided by enabling split
horizon on the serial 0 interfaces.
- Split horizon rule states that router never sends out information on=
an
interface that it learned from that interface
- Let’s see!
|
67
|
|
68
|
- Split Horizon Enabled
- As you can see, with split horizon enabled, RTA does not send RTB (o=
ut
s0) information about 10.1.3.0/24 because it learned it from RTB (sa=
me s0),
and RTB does not send RTA (out s0) information about 10.1.1.0/24 to =
RTA
because it learned it from RTA (same s0). (This also includes the co=
mmon
network between them.
|
69
|
- Split Horizon Enabled – 10.1.3.0/24 down
- RTB notices 10.1.3.0/24 is down and puts this route into hold-down s=
tate
in its routing table. (hold-down coming next)
- RTB immediately sends out a triggered update for only this route (if
there were others in the routing table) with a metric of infinity, 1=
6.
- RTA receives the triggered update and puts the route for 10.1.3.0/24
into hold-down state.
|
70
|
- Split Horizon Enabled – 10.1.3.0/24 down
- Notice that RTA never sends RTB a routing update for 10.1.3.0/24,
because split horizon is enabled on these interfaces.
|
71
|
- Split Horizon with Poison Reverse
- Many vendor implementations of distance vector routing protocols like
Cisco’s RIP and IGRP apply a special kind of split horizon, ca=
lled
split horizon with poison reverse.
- “Split horizon with poison reverse means that, instead of not
advertising routes to the source, routes are advertised back to the
source with a metric of 16, which will make the source router ignore=
the
route. It is perceived=
that
explicitly telling a router to ignore a route is better than not tel=
ling
it about the route in the first place.” (Lewis, Cisco TCP/IP
Routing)
- One drawback is that routing update packet sizes will be increased w=
hen
using Poison Reverse, since they now include these routes.
|
72
|
- Split Horizon Enabled by Default
- Split horizon with poison reverse is enabled by default for all
interfaces except:
- Physical interfaces or multipoint sub-interfaces using Frame Relay or
SMDS encapsulation (CCNA Semester 4 and CCNP Remote Access)
- To disable split horizon on an interface:
-
Router(config-if)# no ip split-horizon
- To enable split horizon on an interface:
-
Router(config-if)# ip split-horizon
|
73
|
- Holddown timers
- The main function of holddown timers is to prevent the distance vect=
or
routing protocol from establishing routing loops during periods of
network transition (topology changes).
- “The rule: Once a
route is marked unreachable, it must stay in this state for a period=
of
time assumed sufficient for all routers to receive new information a=
bout
the unreachable network. In
essence, we instruct the routers to let the rumors calm down and the=
n to
pick up the truth.” (Zinin, Cisco IP Routing)
- The amount of time a router remains in “this state” is
determined by the holddown timer.
|
74
|
- Cisco on-line curriculum information
- You can avoid the count-to-infinity problem by using hold-down timer=
s.
- When a router receives an update from a neighbor indicating that a
previously accessible network is now inaccessible, the router marks =
the
route as inaccessible and starts a hold-down timer.
|
75
|
- Cisco on-line curriculum information (continued)
- Same Route from same neighbor:
Network is back up (Correct News)
- If at any time before the hold-down timer expires an update is recei=
ved
from the same neighbor indicating that the network is again accessib=
le,
the router marks the network as accessible and removes the hold-down
timer.
|
76
|
- Cisco on-line curriculum information (continued)
- Better Route from different neighbor (Correct News)
- If at any time before the hold-down timer expires an update arrives =
from
a different neighboring router with a better metric than originally
recorded for the network, the router marks the network as accessible=
and
removes the hold-down timer.
|
77
|
- Cisco on-line curriculum information (continued)
- Poorer Route from a different neighbor. (Incorrect News)
- If at any time before the hold-down timer expires an update arrives =
from
a different neighboring router with a poorer metric than originally
recorded for the network the update is ignored and the hold-down tim=
er
continues.
- Ignoring an update with a poorer metric when a hold-down is in effect
allows more time for the knowledge of a disruptive change to propaga=
te
through the entire network.
|
78
|
- Additional Information on Holddown Timers
- Flapping routes
- Holddown timers not only help prevent routing loops during transient
periods but also help network stability by dampening unstable, flapp=
ing
routes (routes which continuously go up and down).
- Holddown Time
- As we will see with both RIP and IGRP, the amount of time the router
remains in the holddown state can be modified (with caution!), even =
set
to 0.
- We will look at this later in the presentations on RIP and IGRP.
|
79
|
- Additional Information on Holddown Timers
- Packet forwarding
- Even though routing tables remain constant and routers do not accept
potentially bad updates, an interesting question is whether or not
routers should continue use the existing routes that are in holddown
state for forwarding packets?
- “In practice, routes in the holddown state are used for packet
forwarding.
- In the worst case, packets are forwarded toward the router that was
previously connected to the destination network, which drops them.=
li>
- In the best case, they are forwarded along a potentially suboptimal =
but
valid path.” (Zi=
nin,
Cisco IP Routing)
|
80
|
- Let’s look at a related item in IP, the TTL field.
|
81
|
- When a packet is first generated a value is entered into the TTL fie=
ld.
- Originally, the TTL field was the number of seconds, but this was
difficult to implement and rarely supported.
- Now, the TTL is now set to a specific value which is then decremente=
d by
each router.
|
82
|
- If the router decrements the TTL field to 0, it will then drop the
packet (unless the packet is destined specifically for the router, I=
.e.
ping, telnet, etc.).
- Common operating system TTL values are:
- UNIX: 255
- Linux: 64 or 255 depending upon vendor and version
- Microsoft Windows 95: 32
- Other Microsoft Windows operating systems: 128
|
83
|
- http://www.switch.ch/docs/ttl_default.html
- TTL Overview - Disclaimer:
- The following list is a best effort overview of some widely used TCP=
/IP
stacks. The information was provided by vendors and many helpful sys=
tem
administrators. We would like to thank all these contributors for th=
eir
precious help ! SWITCH cannot, however, take any responsibility that=
the
provided information is correct. Furthermore, SWITCH cannot be made
liable for any damage that may arise by the use of this information.=
- +--------------------+-------+---------+---------+
- | OS Version &nb=
sp;
|"safe" | tcp_ttl | udp_ttl |
- +--------------------+-------+---------+---------+
- AIX n 60 30
- DEC Pathworks V5 n 30 30
- FreeBSD 2.1R y 64 64
- HP/UX 9.0x n 30 30
- HP/UX 10.01 y 64 64
- Irix 5.3 y 60 60
- Irix 6.x y 60 60
- Linux y 64 64
- MacOS/MacTCP 2.0.x <=
span
style=3D'mso-spacerun:yes'> y 60 60
- OS/2 TCP/IP 3.0 y 64 64
- OSF/1 V3.2A n 60 30
- Solaris 2.x y 255 255
- SunOS 4.1.3/4.1.4
y 60 60
- Ultrix V4.1/V4.2A
n 60 30
- VMS/Multinet y 64 64
- VMS/TCPware y 60 64
- VMS/Wollongong
1.1.1.1
n 128 30=
- VMS/UCX (latest
rel.) y 128 128
- MS WfW n 32 32
- MS Windows 95 n 32 32
- MS Windows NT 3.51
n 32 32=
- MS Windows NT 4.0  =
;
y 128 128
|
84
|
- The idea behind the TTL field is that IP packets can not travel arou=
nd
the Internet forever, from router to router.
- Eventually, the packet’s TTL which reach 0 and be dropped by t=
he
router, even if there is a routing loop somewhere in the network.
|
85
|
- Instructors:
- If you have any questions or comments, please email:
- Rick Graziani
- Cabrillo College
- graziani@cabrillo.cc.ca.us
|
86
|
- Ch. 11 – Routing Theory – End of Part 2
- CCNA Semester 2
- Rick Graziani, Instructor
|