eigrp or ospf which should i use

© 2010 Cisco Systems, Inc. All rights reserved.Cisco Public

1

Kevin Delgadillo, PLM, IP Routing, NSSTG

Ernie Mikulic, PM, OSPF, PfR, SAF

EIGRP or OSPF – Which should I use?

© 2008 Cisco Systems, Inc. All rights reserved. Cisco Public 2

Which routing protocol is better?

Which routing protocol should I use in my network?

Should I switch from the one I’m using?

2© 2005 Cisco Systems, Inc. All rights reserved.

RST-3210

11048_05_2005_X2

IPv4 EndsMergeIPv6


The Questions

Is one routing protocol better than any other protocol?

Define “Better!”

Converges faster?

Uses less resources?

Easier to troubleshoot?

Easier to configure?

Scales to a larger number of routers, routes, or neighbors?

More flexible?

…

Both are good choices

Cisco offers full-featured implementations of both today

Cisco EIGRP/OSPF deployment in the enterprise is ~50/50 today


The Questions

The answer is yes if:

The network is complex enough to “bring out” a protocol’s specific advantages

You can define a specific feature (or set of features) that will benefit your network tremendously…


The Questions

But, then again, the answer is no!

Every protocol has some features and not others, different scaling properties, etc.

Let’s consider some specific topics for each protocol....


EIGRP or OSPF: Which Routing Protocol?

Link State & Distance Vector

Convergence Speed

Topology and Heirarchy

Summary


Link State & Distance Vector

Link state

• OSPF is an example

• Each router tells the world about its neighbors

• All information passed is connectivity related

• Each node in the network constructs a connectivity map of the network

• Each node keeps identical link-state database from which routing table is derived

• More complex than distance vector protocols

Distance vector

• EIGRP is an example (but does not behave like a “pure” DV protocol)

• Each router tells its neighbors about its world

• Each node shares its routing table with its neighbors

• Simpler than link state protocols


Convergence Speed

Equal Cost Convergence

OSPF Convergence

EIGRP Convergence

Convergence Summary


Convergence Speed

Which protocol converges faster?

OSPF verses EIGRP

Is DUAL faster, or Dijkstra SPF?

Rules of Thumb

The more routers involved in convergence, the slower convergence will be

The more routes involved in convergence, the slower convergence will be


Convergence Speed

Three steps to convergence

Detect the failure

Calculate new routes around the topology change

Add changed routing information to the routing table

The first and third steps are similar for any routing protocol, so we’ll focus on the second step


A

B

C D

F

E

Equal Cost

Start with B>C>E and B>D>E being equal cost

If C fails, B and E can shift from sharing traffic between C and D to sending traffic to D only

Number of routers involved in convergence: 2 (B and E)

Convergence time is in the milliseconds


A

B

C D

F

E

OSPF

C fails

B and E flood new topology information

All routers run SPF to calculate new shortest paths through the network

B and E change their routing tables to reflect the changed topology


SPF

SPF


OSPF

Within a single flooding domain (OSPF area)

Convergence time depends on flooding timers, SPF timers, and number of nodes/leaves in the SPF tree

What happens when we cross a flooding domain boundary?


OSPF

E floods topology changes to C and D

C and D summarize these topology changes and flood it to B

B builds a summary from the summary flooded to B, and floods it into area 2

A calculates a route to B, then recurses C onto B

Convergence time is dependent on the network design

A

B

C D

F

E

Area 1

Area 0

Area 2


OSPF– Convergence Data

Convergence time with default timers and tuned timers

IPv4 and IPv6 IGP convergence times are similar

- The IPv6 IGP implementations

might not be fully optimized yet- Not all Fast Convergence optimizations might be available

0.000

0.500

1.000

1.500

2.000

2.500

0 500 1000 1500 2000 2500 3000

Number of Prefixes

Tim

e

IPv4 OSPF

IPv6 OSPF

Linear (IPv4

OSPF)Linear (IPv6

OSPF)

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45

0.5

0 500 1000 1500 2000 2500 3000

Number of Prefixes

Tim

e

IPv4 OSPF

IPv6 OSPF

Linear (IPv6

OSPF)Linear (IPv4

OSPF)

All specifications subject to change without notice


OSPF

Within a flooding domain

The average convergence time, with default timers, is on the order of seconds

With optimal SPF/LSA timers, the convergence time can be in the milliseconds

Outside the flooding domain

Network design and route aggregation are the primary determining factors of convergence speed


A

B

C D

F

E

EIGRP

DUAL works on a simple geometric principle:

If my neighbor’s cost (RD) to reach a given destination is less than my best cost (FD), then the alternate path (FS) cannot be a loop

B>D>E>F is 35

B>C>E>F is 30

D>E>F is 20, which is less than the best path, 30, so B>D>E>F cannot be a loop

FC Rule: Choose FS for path where RD<FD

10

10 15

10 10

10

30 35

20

FD = Feasible Distance

RD = Reported Distance

FC = Feasibility Condition

FS = Feasible Successor


A

B

C D

F

E

EIGRP

B will install the path through C, and mark the path through D as a Feasible Successor (FS) in the topology table

When C fails, B looks for alternate loop free paths (FS)

Finding one, it installs it

Local repair, no flooding

Convergence time is in the milliseconds


10

10 15

10 10

10


A

B

C D

F

E

EIGRP

If the second path cannot be proven loop free

B and E detect the failure, and have no alternate path

B queries A and D

A replies that it has no path

D replies with its alternate path

E queries D and F

F replies that it has no path

D replies with its alternate path

Hop-by-hop queries; no flooding


EIGRP

For paths with feasible successors, convergence time is in the milliseconds

The existence of feasible successors is dependent on the network design

For paths without feasible successors, convergence time is dependent on the number of routers that have to handle and reply to the query

Query range is dependent on network design

Good design is the key to fast convergence in an EIGRP network


Convergence Summary

We can sort typical convergence times into three groups:

EIGRP with a feasible successor

OSPF with modified SPF/LSA throttle timers

EIGRP without a feasible successor and good design

OSPF with default timers

EIGRP without a feasible successor without good design

Good

Best


Convergence Summary

It’s possible to converge in under one second using either protocol, with the right network design

Rules of Thumb:

More aggregation tends towards better performance for EIGRP

Less aggregation tends towards better performance for OSPF

If you’re going to use OSPF, tune the SPF/LSA timers


Topology

Hub and Spoke

Full Mesh

Support for Hierarchy

Topology Summary


OSPF Hub and Spoke

OSPF relies on every router within a flooding domain to have the exact same view of the network’s topology (link state database) to calculate loop free paths

OSPF flooding rules have implications for scaling and design in hub and spoke networks


OSPF Hub and Spoke

Although B can only reach C through A, it still receives all of C’s routing information

As the number of remote sites increases, the amount of information each remote site must process and store also increases

This limits scaling in link state hub and spoke networks

B

A

C

D

reachability

only

through A

all link state

information

is flooded

to B


OSPF Hub and Spoke

Controlling route distribution

There’s no way to allow C and D to receive information about 10.1.1.0/24, and not E and F

BA

10.1

.1.0

/24

C

D

E

F

Area 0

Area 1


EIGRP Hub and Spoke

Controlling query range

If A loses its connection to 10.1.1.0/24, it builds and transmits five queries: one to each remote, and one to B

Each of the remote sites will query B

B must process and reply to five queries

BA

10.1

.1.0

/24


EIGRP Hub and Spoke

If these spokes are remotes sites, they have two connections for resiliency, not so they can transit traffic between A and B

A should never use the spokes as a path to anything, so there’s no reason to learn about, or query for, routes through these spokes

BA

10.1

.1.0

/24

Don’t Use

These Paths


EIGRP Hub and Spoke

To signal A and B that the paths through the spokes should not be used, the spoke routers can be configured as EIGRP stubs

BA

10.1

.1.0

/24

router#config t#

router(config)#router eigrp 100

router(config-router)#EIGRP stub connected

router(config-router)#


EIGRP Hub and Spoke

Marking the spokes as stubs allows them to signal A and B that they are not valid transit paths

A simply will not query the remotes, reducing the total number of queries in this example to 1

BA

10.1

.1.0

/24


EIGRP Hub and Spoke

Marking these remotes as stubs also reduces the topological complexity (meshiness) of the network

Without stub configuration on spokes, B believes it has five paths to 10.1.1.0/24, so it has to maintain five topology table entries

BA

10.1

.1.0

/24


EIGRP Hub and Spoke

Routers which are configured as EIGRP stubs will only advertise locally connected or redistributed destinations

These remotes will not pass A’s advertisement of 10.1.1.0/24 to B

B will only have one path to 10.1.1.0/24

BA

10.1

.1.0

/24


Full Mesh

Full mesh topologies are complex:

2 routers = 1 link

3 routers = 3 links

4 routers = 6 links

5 routers = 10 links

6 routers = 15 links

…

Adjacencies = links(links-1)/2


OSPF Full Mesh

Flooding routing information through a full mesh topology is also complicated

Each router will, with optimal timing, receive at least one copy of every new piece of information from each neighbor on the full mesh

OSPF uses notion of Designated Router (DR) to improve scalability in mesh networks

New Information


EIGRP Full Mesh

Routes must be advertised between every pair of peers in the mesh so each router has the correct next hop and routing information

Number the links so they can be summarized to a single advertisement at the edge

Good for smaller mesh networks, summarization more important for larger mesh networks

Summarize

Summarize

Summarize

Summarize

Summarize

Summarize


OSPF Support for Hierarchy

OSPF requires a hierarchical design

Summarization and filtering occur at flooding domain borders

Summarization and filtering can also be configured at routers redistributing routes into OSPF

In a two layer hierarchy, the flooding domain border naturally lies on the aggregation/core boundary

area 0

Su

mm

ari

za

tio

n


EIGRP Support for Hierarchy

EIGRP does not require a heirarchical design

Auto-summarization enabled by default at classful network boundaries

EIGRP enables you to summarize at any desired boundary

Proper network design is still needed!

Distribution

Access

Core

Su

mm

ari

zati

on


Topology Summary

Rules of Thumb

EIGRP performs better in large scale hub and spoke environments

OSPF perform better in large full mesh environments, if tuned correctly

EIGRP tends to perform better in more strongly hierarchical network models, OSPF in flatter networks


Other Considerations - 1

EIGRP forms adjacencies and exchanges routing updates with neighbors

OSPF forms adjacencies with DR/BDR

OSPF can be more efficient than EIGRP for large meshed networks

EIGRP uses metric based on bandwidth and delay

OSPF uses interface cost (inversely proportional to bandwidth)

EIGRP may provide more flexibility in selecting best path

EIGRP by default limits usage to at most 50% of link bandwidth in worst case

OSPF uses 100% of link bandwidth when required

EIGRP may be better suited for lower bandwidth WAN applications

EIGRP provides feature velocity, but is Cisco-proprietary

OSPF is an Internet RFC standard


Other Considerations - 2

EIGRP sends hop-by-hop queries only when Feasible Successor cannot be found

OSPF regularly syncs LSA database and floods network with topology change

EIGRP can be more efficient by minimizing routing information exchanged

EIGRP is a conceptually simpler routing protocol

OSPF’s rules for different types of areas and LSAs can be conceptually more difficult to understand

Some customers believe EIGRP is easier to implement, but both are feature-rich and scalable

EIGRP supports automatic summarization

OSPF’s requires manual summarization

Care is needed in either case to ensure proper summarization!

EIGRP supports both equal and unequal cost load sharing

OSPF only supports equal cost load sharing


Summary


Which routing protocol is better?

Which routing protocol should I use in my network?

Should I switch from the one I’m using?

Did we answer these questions???

42© 2005 Cisco Systems, Inc. All rights reserved.

RST-3210

11048_05_2005_X2

IPv4 EndsMergeIPv6


Summary

There is no “right” answer!

“IT DEPENDS…”

Consider:

Your business requirements

Your network design & topology

Convergence time requirements dictated by your applications

Other intangible factors

EIGRP and OSPF are generally pretty close in capabilities and development (GR, BFD, IPv4/IPv6)


Summary

EIGRP

LargeMesh

Hub and Spoke

Flat Aggregated

Flat Hierarchical

OSPF

Rules of Thumb

Complex Simpler

eigrp or ospf which should i use

Documents