EECS 228 Abhay Parekh [email protected]/228a02/Lecture...EECS 228 Abhay Parekh...

Hierarchical Routing

EECS 228Abhay [email protected]

October 16, 2002 Abhay K. Parekh: Topics in Routing 2

Hierarchical Routing

� Is a natural way for routing to scale� Size� Network Administration� Governance

� Exploits address aggregation and allocation

� Allows multiple metrics at different levels of the hierarchy

6

4

3

2

13

2

4

3

6

13

7

8

5

1 12

1011

Inter DomainRouting OSPF

RIP

IGRP


Why is hierarchical routing important?

� The internet is an interconnection of unequal networks

� Interconnection arrangements drive � the competitive landscape� the robustness of the network� end-to-end performance

� Interconnection is central to all large networks� Voice� Data� Wireless� Cable


Why are there so many players?

� www.thelist.com� How many ISP�s in the 415 area code?

� That start with A-C: about 200�� Just DSL that start with A-C: about 80

� In the telephone network� How many independent telephone companies in

1894-1902 in the US?� 3039 commercial companies, 979 co-operatives

� By controlling interconnection Bell got rid of them� Interconnection is now regulated (CLECs)


What is an Interconnection

� Access to sites reachable via routing and transport facilities� But could also include:

� Wire� SLA + Lease

� Space� Size� Space

� Access to OSS� Dispute Resolution Process� Term


Interconnections occur at many levels

6

4

3

2

13

A

B

C

2

4

3

6

13

7

8

5

1 12

1011

Inter DomainRouting OSPF

RIP

IGRP



6

4

3

2

13

A

B

C

2

4

3

6

13

7

8

5

1 12

1011

OSPF

RIP

IGRP

A.1

D.1

D.2

D.3



6

4

3

2

13

A

B

C

2

4

3

6

13

7

8

5

1 12

1011

OSPF

RIP

IGRP

A.1

D.1

D.2

D.3

E.1

E.2

E.3



1313

75

1 12

11

A.1

D.1

E.2

D.2

D.3

E.1

E.3

D

E


Examples of overlaid interconnecting networks� IP over ATM� Multicast over IP� DSL over POTS� IP over CATV� Etc., � Each involves routing


Two ways to interconnect IP Networks…

� Peering� The business relationship whereby ISPs

reciprocally provide to each other connectivity to each others� transit customers

� Transit� The business relationship whereby one ISP

provides (usually sells) access to all destinations in it�s routing table

William B. Norton, “Internet Service Providers and Peering”


Peering and Transit Figures fromWilliam B. Norton, “Internet Service Providers and Peering”


Benefits of Transit v/s Peering



Moving from Transit to Peering



Peering Attributes

� Bandwidth Pricing: Everything you can think of� Traffic may be asymmetric (web servers)� Clout may vary� Some existing and suggested methods

� Zero Charge (Bill and Keep)� Average Cost� Fully distributed cost pricing� Ramsey Pricing� Wholesale Pricing � Marginal Cost Pricing

� Bandwidth is undifferentiated (can�t peer for video quality bw)� Connection Method

� Direct Connection� Internet Exchange


Internet Infrastructure providesundifferentiated service

� More capacity is thrown at the undifferentiated network, and emphasis continues on �speeding up the internet�, but this just speeds up existing applications

� No future for internet media or other bandwidth intensive applications� No future for significant high speed access penetration� These are huge lost opportunities!!

Undifferentiated Network here to stay?Undifferentiated Network here to stay?

No BusinessModel Cop-out No way to charge, peer or deliver

high speed/quality sensitive applications


Name of the Game: Reachabilty

� BGP is the way by which ISPs co-operate on reachability

� Routing efficiency and performance is important, but not essential� E.g. Path Vector uses many messages

� Power of BGP is that it can express many different ISP routing policies without exposing internal network statistics such as load and topology

� Tremendous growth in the last 10 years�


Skitter Legend

� Plot the AS based on polar co-ordinates (r,θ):� r = 1- log (As degree +1 / Max Degree+1)

� Higher the degree lower the radius� Θ = longitude of AS headquarters


4/1-4/16 2002�1,224,733 IP addresses, �2,093,194 IP links, �932,000 destinations, �70% of globally routable network prefixes; �10,999 ASes (84% of ASes),�34,209 peering sessions


BGP

� Runs over TCP port 179� One Border Routers can be involved in multiple

sessions� Border Routers

� from the same AS speak IBGP� from different AS�s speak EBGP

� EBGP and IBGP are essentially the same protocol� IBGP can only propagate routes it has learned directly from

its EBGP neighbors� All routers in the same AS form an IBGP mesh� Important to keep IBGP and EBGP in sync


Four message types

� Open: Session establishment id exchange� Notification: exception driven information� Keep Alive: soft state � Update: path vector information


Update Message

� Withdrawn Routes: No longer valid

� Attributes: Path Vector, weights and other information about each of the destinations

� <length,prefix>: CIDR notation for the destination

Infeasible Route Length

Withdrawn Routes (variable)

Total Path Attribute Len

Path Attributes (variable)

Length|Prefix

<length,prefix>..


Classless Inter-domain Routing Addresses

� 32 bits in the address divided into 4 8-bit parts, A.B.C.D � Each part takes value 0,1,2,�,255

� E.g. 128.23.9.0� Specify a range of addresses by a prefix: X/Y

� The prefix common to the entire range is the first Y bits of X.� X: The first address in the range has prefix X� Y: 232-Y addresses in the range

� Example 128.5.10/23� Common prefix is 23 bits: 01000000 00000101 0000101� Number of addresses: 29 = 512

� Prefix aggregation� 128.5.10/24 and 128.5.11/24 gives 128.5.10/23

� Addresses allocated by central authority: IANA � Routers match to longest prefix


Advertising a prefix

� One router telling another one � The prefix� IP address of the next hop� Path list of AS�s that the announcement has

passed through� Since announcement propagates from destination, this

yields the path

� No refresh messages required� The announcing router will follow the path

itself


Example

6

4

3

2

13

A

B

C

2

4

3

6

13

7

8

5

1 12

1011

OSPF

RIP

IGRPb via A

b

IBGP

IBGP

BGP Sessiona

c

Announce a


Multihoming

� Two or more interdomainconnections between the same AS�s

� Two or more interdomainconnections between a customer and ISPs

6

4

3

2

A

B

2

4

3

6

7

8

5

1

RIP

IGRP

IBGP

IBGP

a


Multiexit Discriminators (MEDs)

� Way for one AS to influence routing decisions of another AS

� AS_A wants to tell AS_B that network a is closer to router 2 than to router 3

� Router 2 advertises a smaller MED value for a than Router 3

� AS_B prefers the path to a that does not go through 6 and 3

� AS_B does not propagate MEDS from AS_A any further

6

4

3

2

A

B

2

4

3

6

7

8

5

1

RIP

IGRP

IBGP

IBGP

a


Local Preference (for IBGP)

� Similar to MEDs but rather than being part of the EBGP announcement, is a way for IBGP within an AS to prefer one path over another for the same prefix

� Example� Choose the slower path when the prefix is to a

competitor�s network


Multihoming

� Advantages� Redundancy� Load Sharing� Lowest Cost Routing� Lowest Latency Routing

� Disadvantages� Aggregation (if one customer has two ISPs)


BGP Policies

� Multiple ways to implement a �policy�� Decide not propagate advertisements

� I�m not carrying your traffic� Use MEDs to improve performance� Decide not to consider MEDs but use shortest hop

� Hot potato routing� Prepend own AS# multiple times to fool BGP into

not thinking AS further away� Many others�


BGP and Performance

� BGP isn�t designed for policy routing not performance� Hot Potato routing is most common but suboptimal� Performance isn�t the greatest

� 20% of internet paths inflated by at least 5 router hops� Very susceptible to router misconfiguration

� Blackholes: announce a route you cannot reach� October 1997 one router brought down the internet for 2 hours

� Flood update messages (don�t store routes, but keep asking your neighbors to clue you in). 3-5 million useless withdrawals!

� In principle, BGP could diverge � Various solutions proposed to limit the set of allowable policies� Focuses on avoiding �policy cycles�


BGP Updates (Labovitz)

� Most updates were bogus withdrawals� This was to a large extent due to bad

implementations


Non-Withdrawal Updates (Labovitz)

� AADup is the advertising of a route that was just withdrawn


BGP Bouncing Problem (Labovitz et al)

� Initially direct route preferred by each node

R

1

0 2




� R has a fault so sends withdrawal messages to each of its neighbors

R

1

0 2





R

1

0 2





� Nodes choose clockwise 2 hop paths

R

1

0 2






� Detecting loops they choose anticlockwise 2 hop paths

R

1

0 2







R

1

0 2







� And so on�� N! steps, and even more

messages

R

1

0 2


Other problems

� Most router messages bogus� Routing table oscillations worse under path

vector when compared to distance vector� Timer interactions with router vendor

implementations create pathologies� Network Administrator configuration errors

can create catastrophic outages throughout the network


Precautions

� Route Damping� Don�t believe updates unless they make sense� Improves oscillations but adds to convergence time

� More state� Keep history of received and sent messages� Improves situation but quite expensive

� Router Configuration Restrictions� Save Path Vector Protocols (Griffin et al)� Human error potential is still considerable

� Alternatives� Detour� Routing Feedback based approaches� Hard to change the installed base!


Conclusions

� Hierarchical Routing is a core component of most large networks

� Internet interconnection is complex and multi-faceted

� The ability for the network to expand its range of services is tied to the modes of peering

� Internet routing is becoming less hierarchical and more driven by policy than performance

� BGP stability is important and a work in progress


References

� Peering� Norton: �Internet Service Providers and Peering�� Eli Noam: �Interconnecting the Network of Networks�, MIT

Press, 2001.� BGP: Basics

� BGP Tutorial by Nina Taft, SprintLabs� Halabi, �Internet Routing Architectures�, Cisco Press 1997

� BGP: Instability� Labovitz et al (see readings)� Griffin et al (see readings)

EECS 228 Abhay Parekh [email protected]/228a02/Lecture...EECS 228 Abhay Parekh...

Documents

Transcript of EECS 228 Abhay Parekh [email protected]/228a02/Lecture...EECS 228 Abhay Parekh...