1 Understanding Route Redistribution ICNP 2007 October 17 th, 2007 Franck Le, Geoffrey G. Xie, Hui...

1

Understanding Route Redistribution

ICNP 2007October 17th, 2007

Franck Le, Geoffrey G. Xie, Hui Zhang

2

Internetwork and Routing

• Common view: – Intra-domain routing using OSPF, RIP– Inter-domain routing using BGP

• In reality, internetworking is much more complex– ISP networks:

• OSPF routes to be redistributed into BGP (and vice versa)

– Enterprise networks: • When BGP is not used, needs mechanism to distribute

routes among OSPF, RIP, EIGRP domains• Also, needs to distribute routes among multiple OSPF

domains

3

What is Route Re-Distribution (RR)?

router ospf 27

redistribute rip metric 200 subnets route-map rip2ospf

distance ospf external 200

!

route-map rip2ospf permit 100

match ip address 100

set tag 22

set metric-type-1

A

B D

E

Office branch 1 Office branch 2

RIP OSPF

RIP OSPF Local

FIB

C

By default, OSPF routers have no visibility of RIP routers

4

How Does RR Compare to BGP?

• In many scenarios, RR, not BGP, is used to interconnect network domains,

• Even when BGP is used, RR is required to connect BGP and IGP

• RR can implement policy, like BGP• Unlike BGP, RR is NOT a protocol

– RR is just a configuration mechanism, used separately at each router

RR is more commonly used than BGP, but much less understood, and much more error-prone

5

Problem Statements

• Given an internetwork with RR configurations, what are the loop-free and convergence properties?

• What are the guidelines of using RR if one wants to have loop-free and convergent internetwork?

6

Synthesis of the Paper

• Model that reasons about the loop-free and convergence properties

• Sufficient condition to guarantee loop-free and convergence properties

7

Outline

1. Introduction to Route Redistribution (RR)

2. Illustration of routing anomalies

3. A Model for RR

4. Sufficient condition for loop-free and convergent RR

8

Route Selection Process

A

B

C

D

E

Office branch 1 Office branch 2RIP OSPF

RIP

FIB

OSPF Local

P

P

P Signaling

Data path

9

Route Selection Process

A

B

C

D

E


RIP

FIB

OSPF Local

Selected routing process

P P

PP

P Signaling

Data path

OSPF110120 0/1

10

FIB

Route Redistribution Process

A

B

C

D

E


RIP OSPF Local110120 0/1OSPF

RIP Update

P

P Signaling

Data path

11

Outline



3. A Model for RR


12

Instabilities

• Wide range of possible routing instabilities

• No general guideline to configure RR

13

RIP

OSPF

RIP

Formation of Routing Loops

A

B

C

D

E

RIP(120) OSPF(110)

OSPF Local

FIB

RIP OSPF Local

FIB

P

Next-hop: B

Next-hop: C

Next-hop: E

Next-hop: D

P

P

P Signaling

Data path

14

Outline



3. A Model for RR


15

Challenges

• Too many network elements– Hundreds or thousands of routers

• Different router processing order – Routers may process signaling messages in

different order (message delay, router load)– Different order can result in different outcome

16

Solutions

• Too many network elements– Abstractions: routing instances– Logics: route selection, RR, network-wide RR

• Different router processing order – Activation sequence1

1 L. Gao and J. Rexford, Stable Internet Routing Without Global Coordination, in Proc. ACM SIGMETRICS, 2000

17

A Model for RR

• Abstracts the dynamic exchange of routing information for a prefix P

• Allows to predict paths

18

Route Propagation Graph

• Routing instance

• Originating routing instance

• Configured redistribution

• Actual redistribution

• Route vs. no route

• Variables: CL, S

2(110)

1(120)

1(120)

2(110)

80, A, 90

1(120)

2(110)

80, A, 90

1(120)

2(110)

80, A, 90 2(110)

19

Illustration of Model

2OSPF1(110)

3RIP

(120)

4OSPF2(110)

F

F

L

L

H

H

E

E

0Local

(0)

1RIP

(120)

A

A

B C D E

F G H I

K L M N

J

RIP RIPOSPF1 OSPF2

P

20

Illustration of Model Sequence 1

Signaling

Data path

1RIP

(120)

2OSPF1(110)

3RIP

(120)

4OSPF2(110)

F

F

L

L

H

H

E

E

0Local

(0)

A1

RIP(120)

2OSPF1(110)

3RIP

(120)

CL(t=0) = {A} CL(t=1) = {E, F} CL(t=2) = {E, L} CL(t=3) = {E, H}

CL(t=4) = {E}CL(t=5) = {A, F}CL(t=6) = { }

S(t=1) = {A} S(t=2) = {F} S(t=3) = {L}

S(t=4) = {H}S(t=5) = {E}S(t=6) = {A, F}

4OSPF2(110)

21

Route Redistribution Configuration - Cycle Detection (RRC-CD) Problem

• Given a RR configuration, determining whether there is an activation sequence such that the redistributions converge to state including a cycle of active redistributions is NP-hard

22

Outline



3. A Model for RR


23

Sufficient condition for safety

• Pruning of Route Propagation Graph– For each redistributing router, only conserve

redistributions from the routing processes with lowest administrative distances

• Rationale– Focus on preferred redistributions

1(100)

2(70)

3(120)

4(90)

A A A

24

Sufficient condition

If resulting graph satisfies1. Every redistributing router redistributes from a

single routing instance (predictable outcome)

2. For all vertice, there is a redistribution path from a originating vertex (active redistribution)

3. The graph is acyclic (no cycle)

Then, the redistributions converge to an acyclic routing state

No route oscillations No forwarding loops

25

Application of Sufficient Condition

1RIP

(120)

2OSPF1(110)

3RIP

(120)

4OSPF2(110)

F

F

L

L

H

H

E

E

0Local

(0)

A

26


Modifications

1RIP

(120)

2OSPF1(110)

3RIP

(120)

4OSPF2(110)

80, F

F, 80

L

L

H

H

80, E

E, 80

0Local

(0)

A

27


Pruning

1RIP

(120)

2OSPF1(110)

3RIP

(120)

4OSPF2(110)

80, F

F, 80

L

L

H

H

80, E

E, 80

0Local

(0)

A

28


Pruning

1RIP

(120)

2OSPF1(110)

3RIP

(120)

4OSPF2(110)

80, F L

H

0Local

(0)

A

80, E

29


1. Every redistributing router is redistributing from a single routing instance.

2. For all vertice, there is a redistribution path from a originating vertex.

3. The graph is acyclic.

1RIP

(120)

2OSPF1(110)

3RIP

(120)

4OSPF2(110)

80, F L

H

0Local

(0)

A

80, E

30

Summary

• Internetwork is far more complex with RR than the conceptual model of BGP/OSPF

• RR serves a fundamental need, but is not well-understood or even well-designed

• First formal study route-free and convergence properties of RR internetwork– Model – Sufficient condition

31

Future Work

• If one were to re-design the RR, what should be the solution that supports all the RR applications but avoid the pitfalls?

1 Understanding Route Redistribution ICNP 2007 October 17 th, 2007 Franck Le, Geoffrey G. Xie, Hui...

Documents

Transcript of 1 Understanding Route Redistribution ICNP 2007 October 17 th, 2007 Franck Le, Geoffrey G. Xie, Hui...