DD2491 p2 2009

Inter-domain routing and BGP

Olof Hagsand KTH/CSC

DD2491, p2 2009
 Literature
• Practical BGP
– Follow reading instructions

• RFC 4271
• Many vendor pages

DD2491, p2 2009
 Inter-domain routing
• The objective of inter-domain routing is to bind together
all the thousands of independent IP networks that
constitute the Internet
• Perspective from one network
– Decide how to receive information (packets) from the outside
world
– Decide how to spread information to the outside world

• Handling of prefixes
– Receive and choose (filter) between prefixes from other
domains
– Announce prefixes to other domains

• Address aggregation
DD2491, p2 2009
 What is BGP?
• Border Gateway Protocol version 4
• Defined in RFC 4271
• An inter-domain routing protocol
• Uses the destination-based forwarding paradigm
– No other relations can be expressed: sources, tos, link load

• Uses path-vector routing

• Views the Internet as a collection autonomous systems
• Exchanges information between peers using TCP as underlying protocol
• Maintains a database (RIBs) of network layer reachability information
(NLRI:s)
• Supports a toolkit of mechanisms to express and enforce policies
decisions at the AS level

DD2491, p2 2009
 IGP/EGP

ISP
Customer
EGP
IGP
IGP

EGP IGP
– Exterior Gateway Protocol. – Interior Gateway Protocol.
– Runs between – Runs within a
networks/domains (inter- network/domain (intra-
domain) domain)
– Examples: BGP, static routing – Examples: RIP, OSPF, IS-IS.
– Note that BGP can also run
internally in a network: IBGP

DD2491, p2 2009
 Why cant we use an IGP?
• On page 3 in the book, there is a chapter:
– Why not use a single protocol for both internal and external
routing?

DD2491, p2 2009
 Autonomous Systems (AS)
• A set of routers that has a single routing policy, that run under
a single technical administration
– A single network or group of networks
– University, business, organization, operator
• This is viewed by the outside world as an Autonomous System
– All interior policies, protocols, etc are hidden within the AS
• Represented in the Internet by an Autonomous System
Number (ASN). 0-65535
– Example: ASN 1653 for SUNET

• Currently, operators are switching to four-byte ASNs

– RFC 4893: BGP Support for Four-octet AS Number Space

DD2491, p2 2009
 General AS graph

AS1 AS2

AS3 AS4 AS5

AS6 AS7 AS8 AS9

DD2491, p2 2009
 Whois example
gelimer.kthnoc.net> whois -h whois.ripe.net AS1653
aut-num: AS1653
as-name: SUNET
descr: SUNET Swedish University Network
import: from AS42 accept AS42
export: to AS42 announce AS-SUNET
import: from AS702 accept AS702:RS-EURO AS702:RS-CUSTOMER
export: to AS702 announce AS-SUNET
import: from AS2603 accept any
export: to AS2603 announce AS-SUNET
import: from AS2831 accept AS2831 AS2832
export: to AS2831 announce any
import: from AS2833 accept AS2833
export: to AS2833 announce any
import: from AS2834 accept AS2834
export: to AS2834 announce any

gelimer.kthnoc.net> whois -h whois.ripe.net AS-SUNET

as-set: AS-SUNET
descr: SUNET AS Macro
descr: ASes served by SUNET
members: AS1653, AS2831, AS2832, AS2833, AS2834, AS2835, AS2837
members: AS2838, AS2839, AS2840, AS2841, AS2842, AS2843, AS2844
members: AS2845, AS2846, AS3224, AS5601, AS8748, AS8973, AS9088
members: AS12384, AS15980, AS16251, AS20513, AS25072, AS28726
members: AS-NETNOD
DD2491, p2 2009 (Edited example)
 ISP Tiers
Tier-1 (TeliaSonera, Sprint, Verizon, NTT, Level3, Google,...)

• Huge ISPs. They do not pay anyone else for transit,

since they exchange traffic (peer) with all other
tier-1 networks. Everyone else pays to peer with
tier-1 ISPs, to get connectivity the whole Internet.
Tier-2
• Large regional ISPs. Do not peer with all tier-1
networks. Buys transit from a few tier-1 network.
Tier-3
• Smaller ISPs. Buys transit form either tier-1 or tier-2
networks. Exchanges traffic at a single IX.
DD2491, p2 2009
 RIPE
• All public IP addresses and AS numbers is provided
by IANA (Internet Assigned Numbers Authority).
www.iana.org.
– IANA also handles the top level domains and protocol
numbers.

• IANA delegate IP blocks and AS numbers to RIRs

(Regional Internet Registry) so they in turn can
delegate space to LIRs (Local Internet Registry).
• RIPE NCC (Réseaux IP Européens Network
Coordination Centre) is the European RIR
– Also handles non European countrys like Israel, Iraq, Iran,
Russia and many more.
– www.ripe.net
DD2491, p2 2009
 RIPE
• RIPE NCC is one of five RIRs in the world
– ARIN, American Registry for Internet Numbers, www.arin.net
– LACNIC, Latin American and Caribbean Internet Addresses
Registry, www.lacnic.net
– APNIC, Asia Pacific Network Information Centre,
www.apnic.net
– AfriNIC, the African Network Information Centre,
www.afrinic.net

DD2491, p2 2009
 RIPE
• There are two “classes” of IP addresses
– Provider Independent (PI)
– Provider Aggregatable (PA)

• PA space is block of prefixes delegated to a

LIR (Local Internet Registry) and to be used by
themselves and their customers.
– If you use PA space and decide to change ISP the address
space have to be returned.

DD2491, p2 2009
 RIPE
• PI space is addresses that end customers can request
directly from RIPE.
– Good for the end customer, but bad for the ISP.
– RIPE will also start charging for PI space

• You have to make a strong case against RIPE to get

the requested amount of IP addresses.
– There are organizations that can help you with the paper
work towards RIPE.

DD2491, p2 2009
 AS graph and peering relations

Tier 1: Full
Internet
connectivity
AS1 AS2
Transit

NSPs Peer
ISPs AS3 AS4 AS5
Customer

Stubs/
Customers AS6 AS7 AS8 AS9

DD2491, p2 2009
 Cost and peering relations

Full You pay for

Internet
connectivity
AS1 AS2 transit traffic.
Transit
-$ -$
NSPs Peer
ISPs AS3 0 AS4 AS5
Customer
+$ +$

Stubs/
Customers AS6 AS7 AS8 AS9

DD2491, p2 2009
 Traffic patterns

Full Note where there

Internet
connectivity
AS1 AS2 are no traffic
arrows!
Transit

NSPs Peer
ISPs AS3 AS4 AS5
Custome
r

Stubs/
Customers AS6 AS7 AS8 AS9

DD2491, p2 2009
 Peering relations
• An abstract way of defining peering relations is for example:
• Prefix sets:
– Define a customer set, a peering set and a transit set

• Example rules:
– Customer prefixes should be announced to transit and peers
– Peer and transit prefixes should be announced to customers
– Prefer prefixes from peers over prefixes from transit
– Do not accept illegal prefixes (RFC 1918 for example), or unknown
prefixes from customers
– Load balance over several transit providers
– Filter traffic (eg src addresses) according to the prefixes announced

DD2491, p2 2009
 Customer / ISP Relations: Stub AS
Announced
Provider
networks,
traffic flows in
def other direction
n1, aul
t
n2
Customer
• Typical customer topology n1, n2

• Can use default route to reach the Provider and Internet

• Customer can use address block of provider
• Customer does not need to be a separate AS
• Typically use static routing but can also use BGP
– Less common: Use a separate IGP (eg RIP) only to
exchange routes between border routers.

DD2491, p2 2009
 Multi-homed customer

ISP ISP ISP

n1, n2

n1, n2

n1, n2
n1, n2
2

n1
, n

,n
n1

2
Customer Customer Customer

• Customer can be multi-homed for reliability and performance reasons

– Load sharing or geographical traffic distribution
• Non Transit AS does now allow external traffic to pass through
• Multi-homed non-transit AS
• You have to think about: How to announce the prefixes? Default
routes? Symmetrical routing? Packet filtering, address aggregation, etc
DD2491, p2 2009
 Customer with multiple providers
ISP1 ISP2

2
n1

,n
,n

n1
2
Customer
n1, n2

• For several upstream providers address aggregation is an

important issue:
• Which address block should the Customer use?
– From ISP1 or ISP2?
– From both?
– Or an independent address block?
DD2491, p2 2009
 Provider: Multi-homed Transit AS

ISP1 ISP2
n3, n4 n5, n6

,n4
n1
,n

,n3
2,
n5

,n2
,n

n1
6
Multi-homed
transit AS
n1, n2

• Transits traffic within own network

• This the most general configuration and is how a
provider works.

DD2491, p2 2009
 BGP sessions
• BGP connections (peerings) are setup manually using TCP
• Two peers must have IP connectivity
• Things to think about (see RTE):
– How are routes imported into AS2?

– How are routes propagated to AS3?

– Which are the BGP nexthops?

– How is external traffic sent through AS2?

AS1 RTD
AS2
RTF RTA

BGP BGP
RTE RTC
AS3
RTB
DD2491, p2 2009
 Example JunOS configuration
routing-options {
autonomous-system 1653
}
protocol bgp {
group external-peers {
type external;
peer-as 42;
neighbor 192.168.200.13;
}
group internal-peers {
type internal;
local-address 192.168.24.1;
neighbor 192.168.16.1;
neighbor 192.168.6.1;
}
AS1653
}
EBGP
IBGP

DD2491, p2 2009
 Path vector protocol
• In a distance-vector protocol, vectors with destination
information are distributed between routers:
• Example:
– <dst: 10.1.10/24, metric: 5, nexthop: 10.2.3.4>

• Distance-vector has problems with converging

– Example: count-to-infinity
• Path-vector extends the information with a path to the destination
– This enables immediate loop detection
– Several other attributes associated with path
• Also, in BGP, the path vector uses AS:s, not IP addresses
– This hides internal structure in the domains
– Loop detection only on AS-numbers!
• Example: <dst: 10.1.10/24, path: AS1:AS3:AS5, nexthop: 10.2.3.4>
DD2491, p2 2009
 BGP Operation

neighbor/peer neighbor/peer
Establishing BGP session (TCP 179)

BGP speaker BGP speaker

OPEN messages

UPDATE messages

KEEPALIVE messages

NOTIFICATION messages

DD2491, p2 2009
 BGP protocol operation
• OPEN messages to initiate a connection and exchange capabilities
• UPDATES contain
– A set of path attributes
– A set of prefixes sharing the path attributes
– A set of withdrawn routes
• BGP compares the AS path and other attributes to select the best path
for a prefix
– Same prefix may be received from several peers
• Path attributes describes properties of the route
– How it was generated, which is the nexthop, various metrics, etc
• NOTIFICATION to signal errors
• KEEPALIVE to check liveness of peer

DD2491, p2 2009
 BGP Connections
• All updates are incremental • This assumes IP connectivity
– Basic case: No refreshes between peers!
– Extension: RFC 2918: Route – But via other mechanism
Refresh Capability fro BGP-4 than BGP

• This assumes reliable delivery! – Either IGP, static, or directly

connected.
• Therefore BGP runs over TCP
• So BGP connections can rely
– Fragmentation,
on an IGP
– Acks,
• BGP does not use the TCP
– Flow control,
keepalives (which by default
– Congestion control is on the order of hours)
– Byte stream
– No automatic neighbour discovery

DD2491, p2 2009
 BGP Finite State Machine
Active Connect

OPEN
Idle
OPEN
TCP connection
NOTIFICATION
NOTIFICATION
OpenSent
NOTIFICATION

KEEPALIVE

OpenConfirm
Established
KEEPALIVE
BGP connection KEEPALIVE UPDATE
NOTIFICATION

DD2491, p2 2009
 BGP message header
• BGP message header format
0 7 1 2 3
5 3 2

Marker

Length Type
• Marker field:
– Authentication of incoming BGP messages
– Detect loss of synchronization between two BGP peers

• Length field: total message length including the header

• Type field: indicates the message type

DD2491, p2 2009
 OPEN message
0 7 1 2 3
5 3 2
Version
My autonomous system

Hold time
BGP identifier
Opt parm len

Optional parameters (variable)

• Version: version of BGP message (current is 4)
• My AS: ASN of the BGP speaker
• Hold Time: Maximum interval between KEEPALIVE or UPDATE
messages
• BGP Identifier: Sender’s BGP ID
• Optional Parameter length: total length of the Optional Parameter
field
• Optional Parameter: use in BGP session negotiation
DD2491, p2 2009
 0
UPDATE message
15

Withdrawn routes length (2bytes)

Withdrawn
routes
Withdrawn routes (variable)

Total path attribute length (2bytes)

Path
Path attributes (variable) attribute

Network Layer Reachability NLRI

Information (variable)

• 3 basic blocks of UPDATE message:

– Withdrawn routes
– Path attributes
– Network Layer Reachability Information (NLRI)

DD2491, p2 2009
 Withdrawn Routes
• Withdrawn Routes Length: total length of Withdrawn Routes
field
– 0 means no routes being withdrawn and Withdrawn Routes field is
not present in this UPDATE message

• Withdrawn Routes field: contains list of prefixes that are being

withdrawn
– Each prefix is encode as a 2-tuple of <length, prefix> (CIDR)

Withdrawn routes length (2bytes)

Withdrawn routes (variable)

DD2491, p2 2009
 Path Attributes
• Total Path Attribute Length: total length of the Path Attribute
field
– '0' indicates that neither NLRI field nor the Path Attribute field is
present in the UPDATE message

• Path Attributes:
– A sequence of path attributes is presents in every UPDATE message
except message that carries only withdrawn routes
– Each part attribute is a triple of
<attribute type, attribute length, attribute value>

Total path attribute length (2bytes)

Path attributes (variable)

DD2491, p2 2009
 Path Attribute type
• Attribute type: consists of type code and flags
• Type code: contains the attribute code maintained by IANA
• Flags
– Bit 0: well-known (0) or optional (1)

• Universally known?

– Bit 1: for optional attribute; non-transitive (0) or transitive (1)

• Pass the attribute to other neighbors?

– Bit 2: for optional transitive attribute; complete (0) or partial (1)

• Is attribute known by all on the path?

– Bit 3: for attribute length; one octet (0) or two octets (1)

– Lower-order four bits: unused and always set to 0

0 1 2 3 4 7 15

Attribute flags Attribute type code

Unused (must be 0)
Optional bit Extended Length bit
Transitive bit Partial bit
DD2491, p2 2009
Network Layer Reachability Information (NLRI)
• Contains list of prefixes that are being advertised
– Each prefix is encoded as a 2-tuple of <length, prefix>
• NLRI length = UPDATE message length – HDR length – Total
length of Path Attributes field – Total length of Withdrawn
Routes field
• NOTE: RFC 4760: Multiprotocol extensions for BGP-4 places a
generalized NLRIs in an NLRI- attribute
– This means that NLRI for non-IPv4 protocols is obsolete!

DD2491, p2 2009
 NOTIFICATION message
0 7 15 23 32

Error Error subcode Data (variable)

• Error code: indication the type of notification

• Error subcode: more specific information about the nature of the error
• Data: contains data relevant to the error e.g. Bad header, illegal ASN
• Data Length = Message Length - 21
0 7 15 23 3
• Example 2
1 2 5000

– Error: 1 Message Header Error

– Error subcode: 2 Bad Message Length

– Data: 5000 is the erroneous length

DD2491, p2 2009
 KEEPALIVE message
• Periodically sent to determine whether peers are
reachable
• Sent at a rate that ensures that hold time will not
expire
– Recommended rate is one-third of the Hold Timer
– Must not be sent more frequently than one per second
– If Hold Timer is 0, periodic KEEPALIVE must not be sent

DD2491, p2 2009
 Path attributes categories
• Path attributes are characterize according to categories:
• Well-known: All BGP implementations must recognize them
– Mandatory: Must always be present in all updates
– Discretionary: May or may not be sent in an UPDATE

• Transitive: Must be passed on to next peer

• Optional: All BGP implementations need not recognize
them
• Optional + Transitive has proven to be an excellent way to
introduce new features seamlessly!
– If a router does not recognize it, it just passes it along to its peers.
– Therefore, almost all novel attributes are optional + transitive

DD2491, p2 2009
 Early BGP path attributes
• Type code 1: ORIGIN (RFC4271)
• Type code 2: AS_PATH (RFC4271)
• Type code 3: NEXT_HOP (RFC4271)
• Type code 4: MULTI_EXIT_DISC (RFC4271)
• Type code 5: LOCAL_PREF (RFC4271)
• Type code 6: ATOMIC_AGGREGATE (RFC4271)
• Type code 7: AGGREGATOR (RFC4271)
• Type code 8: COMMUNITY (RFC1997)

DD2491, p2 2009
 ORIGIN
• Well-known mandatory
• Defines the origin of the path information
• Types:
– IGP (0) NLRI is internal to the originating AS
(eg learnt via IGP)
– INCOMPLETE (2) NLRI is learned by some other means
(eg static route)

• BGP prefers the path with the lowest origin type

• In JunOS, for example, ORIGINs are 0 by default

DD2491, p2 2009
 ORIGIN example

R1 export direct
R2 export static R4 export IGP EBGP
AS2
IBG
RTD RTB
P R1 origin: INCOMPLETE
R2 origin: INCOMPLETE
AS1 R3 origin: (defined by origin)
R3 learnt via EBPG R4 origin: IGP

R1-R4 are routes imported into BGP in four different ways

– R1/R2: Direct / static
– R3: EBGP
– R4: IGP
DD2491, p2 2009
 AS_PATH
• Well-known mandatory
• Contains a sequence of AS path segments
<path segment type, path segment length, path segment value>
– AS_SET (1): unordered set of ASes a route traversed
– AS_SEQUENCE (2): ordered set of ASes a route traversed

• A BGP speaker prepends its ASN to the AS_PATH list when

sending routes to external BGP peers (not to internal
peers)
• Loop detection
• Shorter AS_PATH is preferred

DD2491, p2 2009
 AS_PATH (cont.)

AS2
1 21

AS1
AS3
192.16.1.0/24
X 1
4
3 32
2
1

AS4

• The AS-PATH is used to break loops (between AS:s)

• AS1 announces 192.16.1.0/24 to AS2 and detects its
own ASN when received from AS4
DD2491, p2 2009
 AS SET
AS1 192.16.2.0/24
192.16.2.0/24 AS Path: 1

192.16.2.0/23
AS Path: 3 {1 2}

AS3 AS4
192.16.3.0/24
AS Path: 2

AS2
192.16.3.0/24

• As alternative to sequences of AS:s, a set denotes a set of AS:s.

• Useful in aggregation at AS-level
• Necessary to detect loops

DD2491, p2 2009
 AS_PATH Manipulation
• The AS_PATH can be manipulated to affect inter-domain
routing behavior
• The AS_PATH can be lengthened to make a path less
preferable
• This affects all ASes that receives this prefix update
• Unlike the MED that only can affect how a neighboring
AS sends traffic to you
• Affects how incoming traffic is routed
• Is achieved by prepending dummy ASNs to the AS_PATH

DD2491, p2 2009
 Routing case before Manipulation
192.16.0.0/24 – 30 20 10
AS20 10G AS30
10G
192.16.0.0/24 – 20 10
10G
AS10 192.16.0.0/24 - 10 IX
10G
192.16.0.0/24
1G AS40
192.16.0.0/24 – 40 10

• AS10 has two links to the rest of the world

– 10Gbit/s and 1Gbit/s
• In this case, traffic from the Internet Exchange IX will follow
shortest AS_PATH
– Traffic will thus use the incoming 1G link
• How can AS10 steer traffic from the IX to take the 10G link
instead?

DD2491, p2 2009
 Routing case after manipulation
192.16.0.0/24 – 30 20 10
AS2 10G AS3
10G
0 0
192.16.0.0/24 – 20 10
10G
AS1 192.16.0.0/24 - 10 IX
0
192.16.0.0/24 10G
1G AS4
192.16.0.0/24 – 40 10 10 10
192.16.0.0/24 – 10 10 10 0

• To force incoming traffic to go through the 10G link, AS10

manipulates the AS_PATH
– Insert dummy AS numbers when sending UPDATEs to AS40
– Make the AS_PATH over the 10G link become the shorter one

• Best practice: bogus AS number should be duplicate of own

– Otherwise, the number can be misleading and cause routing loops
DD2491, p2 2009
 NEXT_HOP
• Well-known mandatory
• Defined IP address of the router that should be used as
the next hop to the destinations listed in NLRI
• Next-hop concept for BGP
– External peer: IP address of the peer that announced the
route
– Internal peer:
• Locally originated routes: IP address of the peer that announced
the route
• Routes learned from external: IP address of the external peer
from which the route was learned

– Route on multiaccess medium: IP address of interface of the

router connected to the medium that originated the route
DD2491, p2 2009
 NEXT_HOP (cont.)

192.16.0.0/24
RTC
AS1
1.1.1.1 RTA
3.3.3.3 192.16.1.0/24

EBGP
2.2.2.2

RTD IBG RTB

P
AS2

192.16.0.0/24 via 1.1.1.1 192.16.0.0/24 via

1.1.1.1
192.16.1.0/24 via 2.2.2.2

DD2491, p2 2009
 NEXT_HOP on Multiaccess Media
• When advertising route on a multi-access media , the
next hop can be an IP address of the interface of the
router connected to the medium that originated the
route
– This is called third-party next-hop
10.0.0.0/24

.2
EBGP
OSP
.1 F
.3
11.0.0.0/24
via 10.0.0.3 11.0.0.0/24

DD2491, p2 2009
 MULTI-EXIT-DISCRIMINATOR (MED)
• Optional non-transitive
• Used on external links to discriminate among
multiple links to the same neighboring AS
• Lower MED is preferred
• MED received from external peer must not be
propagated to other neighboring AS:s

DD2491, p2 2009
 MULTI-EXIT-DISCRIMINATOR (cont.)
RTA RTB
AS1

Link A Link B
MED=70 MED=120

RTD
RTE
AS2 RTC
192.16.0.0/24

• AS2 prefers to receive traffic to 192.16.0.0/24 on link A and use

link B as backup
• RTC announces the prefix with low MED, RTE announces the
prefix with high MED.
DD2491, p2 2009
 MULTI-EXIT-DISCRIMINATOR (cont.)
RTA RTB
AS1

192.16.0.0/24 MED=70 Link A Link B 192.16.0.0/24 MED=120

192.16.1.0/24 MED=120 192.16.1.0/24 MED=70

RTD
RTE
AS2 RTC
192.16.0.0/24 192.16.1.0/24

• AS2 wishes to load-balance traffic using MED.

• Receive traffic to 192.16.0.0/24 on Link A
• Receive traffic to 192.16.1.0/24 on Link B

DD2491, p2 2009
 MULTI-EXIT-DISCRIMINATOR (cont.)
RTA RTD
AS1 MED=5
0 AS3

MED=7 MED=120
RTE
0
AS4

RTB RTC
192.16.0.0/24
AS2

• AS1 will select RTB over RTC, but chooses between RTB and
RTD using other means
• MEDs can not be compared from different ASs!
DD2491, p2 2009
 Using MED as tie-breaker
• The use of MED as tie-breaker is controlled by several sub-
settings
• CISCO for example, has a non-deterministic comparison by
default based on age of the routes (newer routes are pairwis
compared).
– cisco-non-deterministic parameter in JunOS.
– deterministic-med in CISCO
• You can also set always-use-med to use MED comparisons
from different AS:s
– Can be useful if you are among a group of AS:s that trust each
other.

DD2491, p2 2009
 LOCAL-PREF
• Well-known discretionary
• Used as local policy to set degree of preference of
routes when announcing to other internal peers
• Used locally within the AS
• A higher local preference is preferred(!)

DD2491, p2 2009
 LOCAL_PREF (cont.)
AS1
192.16.0.0/24

AS2 AS3

T1 Link T3 Link

IBGP

RTB RTC
AS4
Set LOCAL_PREF = 200 Set LOCAL_PREF = 300

• AS4 prefers to send traffic to 192.16.0.0/24 on the T3 Link.

DD2491, p2 2009
 MED versus LOCAL_PREF
• MED is announced to other AS:s
– Used by your neighbors to tell you how they want to receive
traffic

• LOCAL_PREF announced internally

– Used by you to steer traffic internally (to your neighbors)

• LOCAL_PREF overrides MED

• Lower MED preferred
• Higher LOCAL_PREF preferred

DD2491, p2 2009
 ATOMIC_AGGREGATE
• Well-known discretionary
• Set to indicate information loss
– There may be longer prefixes to AS:s not in AS_PATH
– Alternative to using AS_SET

• Receiving BGP speaker must not de-aggregate the route

• BGP speaker that receives a route with this attribute needs to
be aware of that actual path to destinations may not be the
path specified in the AS_PATH attribute of the route
• Should not be set when the aggregate carries some extra
information that indication from where the aggregated
information came

DD2491, p2 2009
 AGGREGATOR
• Used in combination with ATOMIC_AGGREGATE
• Optional transitive
• Contains ASN and IP address of BGP speaker that
aggregates the route
AS1
192.16.0.0/24 AS3
RTA 192.16.0.0/23: AS3 AS2

130.247.203.1
AS2
192.16.0.0/23 ATOMIC_AGGREGATE

AGGREGATOR=(3, 130.247.203.1)

Example: ATOMIC_AGGREGATE and AGGREGATOR

DD2491, p2 2009
 BGP Decision Process
1. If next hop inaccessible, ignore route
2. Prefer highest local preference value
3. Prefer shortest AS_PATH
4. Prefer lowest origin type (IGP, EGP, INCOMPLETE)
5. Prefer lowest MED value (if from same AS)
6. Prefer routes from EBGP over IBGP
7. Prefer routes with lowest IGP metric Nexthop
8. Prefer route from peer with lowest router id
9. Prefer route from peer with lowest address

DD2491, p2 2009
 Vendor specific tie-break
• CISCO has several own rules
– Weight (cisco-specific)
– Routes are compared by default pairwise in the order
they arrived (non-deterministic).
Example:
entry1: AS(PATH) 500, med 150, external, routerid 172.16.13.1
entry2: AS(PATH) 100, med 200, external, routerid 1.1.1.1
entry3: AS(PATH) 500, med 100, internal, routerid 172.16.8.4

– Entry3 is chosen. Why?

– Turn this off with bgp deterministic-med

• Juniper is somewhat more standard compliant

DD2491, p2 2009
 Peering sessions
• Neighbor negotiation of IBGP and EBGP are the same
– IBGP peering: within an AS
– EBGP peering: between AS:s
• Two peers must have IP connectivity
– Simple check: they should be able to ping each other
• If EBGP peers are not physically connected: Multihop EBGP

AS1 RTD
AS2
RTF RTA
EBGP
EBGP Multihop
RTE IBGP RTC
Physical
Logical
AS3
This router does not RTB

DD2491, p2 2009
run BGP
 IBGP loopback peering
AS2
RTC
RTB eth0 lo
RTE
eth1

RTD

• IBGP peering is usually done with loopbacks. Why?

– More stable: Not tied to single physical path, if a
link/interface goes down, another route may be chosen.

• But IBGP needs an IGP so that the loopbacks can be

reached
– And TCP connections can be established

DD2491, p2 2009
 NEXTHOP is
192.168.200.2
How do I reach it?
EBGP peering
AS1 DMZ:
IGP .1 192.168.200.0/30 .2 AS2
IBGP
RTC RTA EBGP RTB

• EBGP peering is typically made over a physical directly

connected link
– Demilitarized zone (DMZ) that does not “belong” to either AS
• But if routes learned via EBGP uses the DMZ address as nexthop
– The DMZ must be redistributed via IGP!
– But the DMZ is not really part of the AS,...
• Although the DMZ is usually a part of one of the AS:s
address ranges

DD2491, p2 2009
 EBGP: Next-hop self
NEXTHOP NEXTHOP is
is 10.0.0.1 192.168.200.2
AS1 lo: 10.0.0.1 DMZ:
IGP .1 192.168.200.0/30 .2 AS2
IBGP EBGP
RTC RTA RTB

• Alternative:
– Set next-hop-self
– Announce routes using the loopback address of the border
router as next-hop
– DMZ does not need to be distributed within the AS
• But RTA still uses the directly connected DMZ address

DD2491, p2 2009
 EBGP nexthop: recursive lookup
RTD
AS1 130.2.3.0/24
.1
12.0.0.0/30
lo: 10.0.0.1 DMZ:
IGP 192.168.200.0/30 AS2
IBGP .1 .2
RTC RTA EBGP RTB

RTC:s routing table

alternatives
Route Nexthop Protocol DMZ
130.2.3.0/24 192.168.200.2 IBGP
192.168.200.0/30 12.0.0.1 IGP nexthop
12.0.0.0/30 - direct

Route Nexthop Protocol

130.2.3.0/24 10.0.0.1 IBGP Next-hop
10.0.0.1/32
12.0.0.0/30
12.0.0.1
-
IGP
direct
self

DD2491, p2 2009
Alternative EBGP peering: multi-hop

EBGP Does not speak

AS1 AS2 BGP

RTA RTB

• Problems: EBGP peering dependent on routing

• Nexthop address is in other AS
– You need to learn the remote next-hop address
– And you may be dependent on the other AS:s IGP for
delivery that may cause instabilities

DD2491, p2 2009
 Alternative EBGP peering: redundancy/load
balancing

AS1 EBGP

RTA RTB
AS2

• Peer between loopback address

• Set equal cost to nexthop using IGP or static routes
• Load balance between links
• Or use one link as redundant link

DD2491, p2 2009
 Alternative EBGP peering: redundancy/load
balancing

AS1 AS2

• Peer between loopback address

• Load balancing between several links/redundant link
• But next-hop may now be dependent on IGP/IBGP in other AS

DD2491, p2 2009
 How to transit traffic?

AS1 AS2
RTA EBGP RTB
RTC AS3

RTE EBGP RTF

RTD

• How does RTC and RTD know how to forward transit traffic
between RTA and RTF?
• You cannot use default routes. Why?
• You may use IGP to distribute external routes.
• But most common nowadays is to use IBGP to distribute
external routes internally.
DD2491, p2 2009
 How to transit traffic using IGP

RTC
AS1 IGP AS2
IGP
RTA EBGP RTB IGP
IBGP AS3
IGP
RTE EBGP RTF
RTD

• You can inject all BGP routes into your IGP as external routes
• Scales badly – High memory consumption for the IGP and will
take time to converge
• There is also a problem with synchronization between the IGP
and EBGP
– Can you announce a route even though your IGP has not
converged?
DD2491, p2 2009
Synchronization between IGP and BGP
Is 192.16.124.0/24
RTC
AS1 192.16.124.0/24 IGP AS2
propagated in my AS?

IGP
RTA EBGP RTB IGP
IBGP
192.16.124.0/24 AS3
IGP
192.16.124.0/24
RTE EBGP RTF
RTD

• If an AS provides transit service to another AS, a BGP

speaker should not advertise a route to external peer unless
all routers within an AS learned about that route via IGP
– RTE checks that 192.16.124.0/24 is reachable via IGP before
announcing it to RTF via BGP
• If you ignore this: no synchronization

DD2491, p2 2009
 Using IBGP: Full mesh
RTC
AS1 AS2
RTA EBGP RTB
AS3

RTE EBGP RTF

RTD

• With IBGP all transit routers know all external routes.

– Note that all transit routers need to speak IBGP (what happens if they dont?)

• But loop prevention in BGP is via AS_PATH

– There is no change in AS_PATH between internal peers!

• Loop prevention in IBGP:

– All IBGP speakers are fully meshed

– Never reannounce routes to an IBGP peer learned from another IBGP peer

DD2491, p2 2009
 IBGP full mesh
• So IBGP needs to be fully meshed in order for:
– All internal routes to receive all external routes
– Loop prevention (no difference in AS_PATH)

• The number of TCP connections in an AS:

– n*(n-1)/2

• Not practical for large transit networks, but new routers

are pushing the limit upwards due to higher route-
processor performance
• Two ways to remove this scaling limitation (later lecture):
– Route reflectors (RFC 4456)
– AS confederations (RFC 5065)

DD2491, p2 2009
 Route advertisement rules
• BGP next-hop must be reachable
– Consequence: If IGP fails, BGP route is not announced

• Advertise only active BGP routes to peers

– Juniper specific
– Consequence: If same route from IGP is active, BGP route is
not announced!
– Turn off in JunOS with: advertise inactive

• Never forward IBGP routes to IBGP peers (full mesh)

DD2491, p2 2009
 Private ASes
• Sometimes it is not necessary to have a public ASN
• IANA has reserved the ASN range 64512 – 65535 for
internal use within a system
• Can be used for customers that are single-homed or
multi-homed to the same provider
• Private ASNs must not be announced globally
• Providers must strip private ASNs before announcing the
prefixes on to the rest of the Internet
• Purpose of private AS:s is to conserve AS numbers and
hide networks

DD2491, p2 2009
 Private ASes, cont’d

192.16.0.0/24 - 1

R3 AS7
AS1 R2 AS1
R2

192.16.0.0/24 - 65001 R1 AS65001

192.16.0.0/24

• Prefixes originating from AS65001: AS_PATH of 65001

• AS1 propagates the prefix but strips the private ASN

DD2491, p2 2009
 Extensions
• BGP is under constant development
• New operational problems and new technologies
require extensions to the protocol
• Extensions are introduced, standardized, and
implemented
• Implementation of extensions:
– Negotiated via BGP capabilities when peering is set up.
– Sent as optional transitive attributes and either recognized
or not

DD2491, p2 2009
 Extensions example
• BGP extensions
– BGP communities attribute (RFC1997)
– Route refresh capability (RFC2918)
– BGP multipath (RFC3107)
– Capabilities advertisement (RFC3392)
– BGP route reflection (RFC4456)
– Multi-protocol extensions (RFC4760)
– Graceful restart (RFC4724)
– Four-byte AS (RFC4893)
– Autonomous system confederations (RFC5065)

• TCP extension
– TCP MD5 signature option (RFC2385)
DD2491, p2 2009
 Capabilities Advertisement
• An 'extension to negotiate extensions'
• Announce supported capabilities to the peer with OPEN
message using options parameter
• Some capabilities are (~same as previous slide)
Value Description Reference
0 Reserved RFC3392
1 Multiprotocol Extensions for BGP-4 RFC4760
2 Route Refresh Capability for BGP-4 RFC2918
3 Cooperative Route Filtering Capability
4 Multiple routes to a destination capability RFC3107
5-63 Unassigned
64 Graceful Restart Capability RFC4724
65 Support for 4-octet AS number capability RFC4893
66 Deprecated (2003-03-06)
67 Support for Dynamic Capability (capability specific)
68-127 Unassigned
128-255 Vendor Specific

DD2491, p2 2009
 The COMMUNITY attribute
• RFC 1997 defines a 4-byte COMMUNITY attribute as optional
transitive
• A group of destinations that share some common property
• Used to simplify routing policies based on logical property
rather than IP prefix or ASN
• Format
– First 2-bytes ASN, last 2-bytes defines a value (ASN:value)
– Example 5678:90 (0x162E005A)
• A route can have more than one community attribute
• BGP speaker can add and modify a community attribute before
passing routes on to other peers

DD2491, p2 2009
 NO_EXPORT example

AS1 192.16.0.0/24 NO_EXPORT AS2

192.16.0.0/24 192.16.0.0/23

192.16.0.0/23

192.16.1.0/24 NO_EXPORT
192.16.1.0/24 192.16.0.0/23

• NO_EXPORT is a well-known community (0xFFFF FF01)

– a route should not be advertised to peers outside an AS

• Other well-known:
– NO_ADVERTISE: a route should not be advertised to other BGP
peers
DD2491, p2 2009
 Extended communities
I T Type[Subtype] Data
1 byte 1 byte 6-7 bytes

• Extended communites are 8-bytes with more structure

– RFC 4360

• Type 0: <ASN:2; Data:4>

• Type 1: <IPv4:4; Data:2>
• Type 2: <ASN:4; Data:2>
• Example: Used in VPNs to tag VPN-specific information

DD2491, p2 2009
 Use of Communities
• Communities are used extensively in modern networks for
defining policies
– Both internally and between networks (if they have agreed)

• You “tag” a route with a community and use this information to

implement a policy
– Tag backbone routes
– Tag routes you wish to advertise to peers
– Tag routes with VPN label
– Eg: SUNET uses communities to tag routes with academic and non-
academic sites

• See Practical BGP p 217 ff, for more examples

DD2491, p2 2009
 Community configuration example (1)
Tagging a community at the edge (or by the other peer):

community academic members 3244:11;

policy-statement add-academic {
route-filter 172.16.0.0/8 upto /16 {
community add academic;
next policy;
}
route-filter 192.168.0.0/8 upto /24 {
community add academic;
next policy;
}

}
}

DD2491, p2 2009
 Community configuration example (2)
Using the community to implement a policy:

community academic members [3244:11]; # regexp

policy-statement from-academic {
from {
protocol bgp;
community academic;
}
then
as-path-prepend “201;201”;
}

DD2491, p2 2009
 Multiprotocol extension for BGP-4
• Support routing of other network layer protocols than IPv4
• NLRI and NEXTHOP fields in the UPDATE message are IPv4
specific
• Use a generalized address form using:
• AFI - Address Family Identifier
• SAFI - Subsequent Address Family Identifier

• Two new attributes replace NLRI and NEXTHOP

– Multiprotocol Reachable NLRI (MP_REACH_NLRI)
• Contains NLRI and nexthop
– Multiprotocol Unreachable NLRI (MP_UNREACH_NLRI)

• Used in multicast, IPv6, VPNs, etc.

DD2491, p2 2009
 Multiprotocol extension for BGP-4
• Examples
– IPv4 unicast: AFI=1, SAFI=1
– IPv4 multicast: AFI=1, SAFI=2
– L3VPN: AFI=1, SAFI=128
– IPv6 unicast: AFI=2, SAFI=128
– IPX: AFI=11

DD2491, p2 2009
 Route refresh capability
• By default, BGP has no mechanism to dynamically request for
re-advertisement of routes from a peer
• If a route (its attributes) does not change, a BGP speaker does
not re-announce it
• Therefore, a receiver needs to cache all previous routes
– Even if not required at a specific moment
• This places a lot of load (memory) on the receiver
• Suppose a router changes input policy and does not want to
store all data from all neighbors “just in case”
• With the ROUTE-REFRESH message, a router can request to
get the complete Adj-RIB-Out from a neighboring router

DD2491, p2 2009
 TCP MD5 Signature option
• Provides a mechanism for TCP to carry a digest message in
each TCP segment using a shared secret
– MD5 message digest algorithm
– Verification of authenticy (no encryption)
– manually configured

• Protects tcp header, parts of the ip header (addresses) and the

whole TCP payload (ie whole BGP message)
• Helps BGP protect itself from spoofed TCP segments, TCP SYN/
RST, data injection,
• Problem
– MD5 algorithm has been found to be vulnerable to collision search
attack
– Performance issue from calculating and comparing digests
– Few actually use it (less than 10% - G Huston 2009)
DD2491, p2 2009 Practical BGP: pages 333-336
 Graceful restart
• When a BGP peering goes down (eg TCP connection is reset, closed, or
Keepalive fires) a BGP peer stops forwarding to that peer
– Ie, if the control plane fails, it is seen as a sign that the data-plane
fails
• The other peers switches over the routes to other peers
• Then when the router comes back up (BGP peering established) they
may switch over to the restarted router again
• But most hw routers have separate forwarding and control planes
• So forwarding can (in principle) continue without the control plane
– At least for a limited period of time before forwarding data becomes stale

• Graceful restart:
– Tell your peers that: You are going down but you will be back, please
continue forwarding to me until I am back.

Practical BGP: pages 269-276

DD2491, p2 2009
 Route flap damping
• A route that changes (withdrawn, announce or attribute change)
leads to a new UPDATE message.
• A route that changes too often leads to route flaps that can
ripple through the Internet
• A route can be a bad link that goes up and down, or an instable
routing state
– Some scenarios (clusters of routers) can even magnify flapping

• Only a small percentage of the Internet routes cause the

majority of the flaps
• Flaps cause many UPDATE messages, that causes
recomputation of FIBs that cause change in traffic.

Practical BGP: pages 238-241

DD2491, p2 2009
 Route flap damping (cont)
• Idea is to use history of route to predict future
• Introduce a penalty every time a route changes (eg 1000)
• Decay penalty exponentially using a half-time (eg 15 minutes)
• Stop announcing route using two levels:
– Suppress/Cutoff-threshold to stop announcing (eg 3000)
– Reuse threshold to start announcing (eg 750)

• Note: If one specific route is suppressed, a less specific route

can be used for traffic
• Route flap dampening is most effective if everyone uses it
towards the edges. Why?
– But is usually installed towards upstream to protect from
instable remote prefixes.

DD2491, p2 2009
 Route flap damping example
Penalty (flaps)

Stop announcing

Suppress limit

Reuse limit

Start announcing

Time

DD2491, p2 2009
 Four-byte AS numbers
• 2-byte ASNs are quickly running out
• 4-byte ASNs have been standardized re-using the
AS_PATH and a migration technique using a special 2-byte
ASN: 23456.
• The migration technique maps all 4-byte ASNs to 23456
when NEW speakers (that have four-byte AS capability)
talk to OLD speakers (those that do not have 4-byte AS
capability)

DD2491, p2 2009
 AS numbers in BGP
• Where does BGP carry AS numbers?
– In the UPDATE message (my ASN)
– In the AS_PATH attribute
– In the Aggregator attribute
– In Communities attributes

• A NEW speaker announces 4-byte AS as a capability

– The capabilty also includes myASN.

• NEW speakers use the AS_PATH attribute for 4-byte

ASNs

DD2491, p2 2009
 New attributes
• Two new attributes (optional transitive)
– AS4_PATH
– AS4_AGGREGATOR

• Only used to “tunnel” 4-byte AS information over non

4-byte AS clouds.

DD2491, p2 2009
 NEW speaking with OLD
• NEW converts all 4-byte ASNs in AS_PATH to 23456
• NEW creates the attribute AS4_PATH to “tunnel” the
4-byte AS-path to other NEW speakers.
• When NEW receives a route from OLD with an
AS4_PATH attribute, it constructs a new AS_PATH
replacing all 23456 with the corresponding 4-byte
AS:s in AS4_PATH.

DD2491, p2 2009
 Example
AS70000 AS50 AS100

NEW OLD NEW

AS_PATH: AS_PATH: 23456 AS_PATH:

532 532 50
90000 23456 70000
AS4_PATH: 70000 532
532 90000
90000

• How can loop detection work?

• Because you only make loop detection concerning your own AS.
And that is never AS_TRANS.
• But AS_PATH looks strange: 23456 may appear many times
DD2491, p2 2009