BGP

#

• BGP decides on prefix vector attributes, not anything link related like IGP’s do
• AFI Address Family Identifiers
• SAFI Sub-Address Family Identifiers

Autonomous Systems

#

• Public ASNs 1-64511
• Private 64512-65535
• Now also 4-byte with RFC 4893
  • If you ever see AS 23456, that means somewhere in the line was a 2-byte only BGP speaker
  • Real AS_PATH encoded with optional transitive attributes AS4_AGGREGATOR and AS4_PATH

BGP Peerings

#

• Like IGP, BGP must first find neighbors to exchange routing information with
• Unlike IGP
  • BGP does not have own transport
  • BGP has different types of neighbors
  • BGP neighbors not discoverable by default
    • IGPs typically use multicast to discover each other
  • BGP neighbors do not have to be directly connected

BGP Transport

#

• Uses TCP/179
• BGP neighbor statement tells processes to
  • Listen for remote address on TCP/179
  • Initiate session to remote address on TCP/179
  • If collision, higher router-id becomes TCP client
• TCP server must agree on client’s expected source IP, or will sent RST
• By default, client packets are sourced from outgoing interface in routing table
  • Can modify this with update-source

BGP Messages

#

• Open
• Keepalive
• Update
• Notification

BGP Neighbor States

#

• Idle
  • Device is not initiating any other states
• Connect
  • Waiting for TCP connection to complete
• Active
  • Trying to initiate TCP connection
• OpenSent
  • Has sent Open, waiting for neighbor to send its Open
• OpenConfirm
  • Waiting for neighbor to send its Keepalive
• Established
  • Fully established peering

Path Attributes

#

• Well-knowns
  • All routers must recognize this PA
• Optional
  • Routers may or may not support this PA

Types

#

• Well-known mandatory
  • This Update must contain the PA or the neighbor will be torn down with a notification error message
  • The 3 are AS_PATH, NEXT_HOP, ORIGIN
• Well-known discretionary
  • PA doesn’t have to be present
  • E.g. Local Preference
• Optional transitive
  • Must pass PA to other neighbors
  • E.g. Community
• Optional non-transitive
  • Doesn’t have to pass PA to other neighbors
  • E.g. MED

The Origin Attribute

#

• An attempt to record where a prefix came from
• 3 Options
  • IGP - Via network statement, and thus presumably known via an IGP
  • EGP - Via legacy EGP, so completely deprecated
  • Incomplete - BGP doesn’t know exactly, so likely redistributed
• Used as a path selection consideration

The AS_PATH Attribute

#

• When a prefix is sent via eBGP and leaves an AS, the AS of the sender is prepended
• Origin AS is on far right of list
• When a prefix is sent via iBGP nothing added

The NEXT_HOP Attribute

#

• eBGP prefixes have the NEXT_HOP set to the neighbor sending
• iBGP does not change the NEXT_HOP when sending
  • Update-self, or IGP reachability is used to address this

BGP Weight

#

• Pseudo-proprietary
• Local value assigned to a prefix and not advertised to others
• Value 0 - 65535, with higher = better
• When prefix is locally generated, it gets 32768
  • Otherwise, default weight is 0

BGP Best Path Selection

#

Cisco Best Path Selection

#

• Highest weight
• Highest LOCAL_PREF
• Prefer locally originated
• Shortest AS_PATH
• Lowest origin type
• Lowest MED
• Prefer eBGP over iBGP
• Lowest IGP metric to the BGP NEXT_HOP
• Oldest path
• Lowest Router ID source
• Minimum cluster list length
• Lowest neighbor address

Juniper Best Path Selection

#

• Highest LOCAL_PREF
• Lowest AIGP
• Shortest AS_PATH
• Lowest origin type
• Lowest MED
• Prefer locally originated
• Prefer eBGP over iBGP
• Lowest IGP metric to the BGP NEXT_HOP
• Active path
• Primary router
• Lowest Router ID source
• Minimum cluster list length
• Lowest neighbor address

eBGP Peerings

#

• eBGP checks for TTL = 1
• eBGP checks that neighbors live on same subnet
• Either side can be the initiator from an ephemeral port to TCP 179 destination port
• Neighbors can use IPv6 address for peering, and still share IPv4 NLRI

iBGP Peerings

#

• iBGP split horizon rule states that iBGP learned prefixes are not passed on to other iBGP neighbors
• We often use IGP advertised loopbacks to form iBGP neighbors

eBGP Multihop

#

• In Cisco, when you enable ebgp-multihop, the disable-connected-check is automatically done in the background

Using BGP Authentication

#

• Typically done for eBGP, but perfectly valid for iBGP too

Misc Neighbor Options

#

• log-neighbor-changes is typically a default since neighbor state is critical
• Hard coding the router-id is a good idea

The Cisco Network Command

#

• Method to advertise a prefix which exists in the RT
• Must specify the mask property for non-classful address
• Good idea to just always use it

Redistributing NLRI in Cisco BGP

#