BGP Notes

The main purpose of this article is to summarize the key concepts of BGP for the my CCIE exam preparation. All commands are related to IOS/IOS-XE, unless stated otherwise.

In addition, I strongly recommend the book Troubleshooting BGP: A Practical Guide to Understanding and Troubleshooting BGP (Networking Technology) by Vinit Jain, Brad Edgeworth and some of BGP contents mentioned below are referenced from it.

I plan to write other blog posts about the summary of EIGRP/OSPF/multicast concepts and regularly review them. Hope it might be helpful for someone in the Internet. 😂

BGP (Border Gateway Protocol):

Path vector routing protocol
Used to exchange networks on the Internet
Transported on TCP port 179
ASN (autonomous system number)

2-byte format

Public: 1-64,551
Private: 64,552-65,535

4-byte format

Public
Private: 4,200,000,000–4,294,967,294

AD (administrative distance)

EBGP: 20
IBGP: 200

BGP Sessions:

EBGP

Peers are in different ASNs
TTL = 1

Configure "neighbor x.x.x.x disable-connected-check" for loopback address peering
OR configure "neighbor x.x.x.x ebgp-multihop x"

IBGP

Peers are in a same ASN
TTL = 255
Require IP reachability

AFI (address family identifier) / SAFI (subsequent address family identifier)

In IOS, the IPv4 address-family is activated by default

Disable the IPv4 address-family via "no bgp default ipv4-unicast" and make the router ready for the multi-protocol support

In NX-OS, the address family must be specified for the neighbor.

Neighbor states and the finite state machine (not demonstrated)

Idle
Active
Connect
OpenSent
OpenConfirm
Established

Convergence consideration:

Using BFD

BGP Messages:

Keepalive
Open

BGP identifier (router ID)
Optional (session capabilities, e.g. route refresh)

Update

Advertise NLRIs (prefixes and its associated PAs)
Withdraw NLRIs (prefixes only)
Also act as a keepalive

Notification

BGP Timers:

Keepalive timer: 60 seconds
Hold timer: 180 seconds
MRAI (minimal route advertisement interval)

IBGP: 0 second
EBGP: 30 seconds

NXOS

0 second for IBGP/EBGP

PAs (path attributes; not listed all):

Well-known mandatory

NEXT_HOP

The next hop of the prefixes received from EBGP peer will be modified.
The next hop of the prefixes received from IBGP peer will not be modified.

Router A (send updates) ---- EBGP ---- Router B ---- IBGP ---- Router C
As for updates from the router A, the next-hop reachability check will fail in the router C and the next-hop will show the update source of the router A
Solutions:

Configure "neighbor router_C_x.x.x.x next-hop-self" in the router B
The router B advertises the EBGP peering link, with the router A, into IGP

AS_PATH
ORIGIN

Well-known discretionary
Optional transitive

Community

Similar to route tag but one route can carry multiple communities
IOS and NXOS devices do not advertise the communities by default

Configuration knob: neighbor x.x.x.x send-community

no-export
local-as
no-advertise

RT (route target)

Extended community
For import/export of routing updates in the VPN environment

SOO (site of origin)

Extended community
For the multi-home (PEs) in the VPN environment, PE will reject the BGP route updates with the same SOO as the configured SOO for the site

Optional non-transitive

Loop Prevention

EBGP

Won't accept the route updates, which has its own ASN in AS_PATH, from other EBGP peers

Break the rule via "neighbor x.x.x.x allowas-in"

IBGP

NLRI updates, received from the IBGP peer, must not be sent to the other IBGP peers (split horizon)

Solutions:

Route reflectors
Full mesh of IBGP peers
Confederation
MPLS

Route Advertisement

There are 3 tables maintaining the prefix and PAs for a route

Adj-RIB-in

Contains NLRIs in the original form advertised from BGP peers
Will be purged after all inbound route policies are processed

"neighbor x.x.x.x soft-reconfiguration inbound" keep the raw NLRI in memory and no route refresh will be sent out after "clear ip bgp * soft in"

loc-RIB

local BGP table

Adj-RIB-out

After all outbound route policies are processed, NLRIs are sent to BGP peers,

Network statement (network x.x.x.x mask x.x.x.x)
- Must match the exact prefix in RIB
- Install the route into the BGP table (loc-rib)
- Not enable neighbor discovery for the interface
Only the best route will be advertised to the BGP peers

Best path selection rules:
Wise Lips Lover Will Apply Oral Medication Every Night

Weight

locally significant

Local preference
Locally originated

network > redistribution > aggregation

Length of AS path
Origin

(i>?)

MED
EBGP > IBGP
Next-Hop with the lowest metric
...

Recalculation of the best path for a prefix:

BGP next-hop reachability change
Failure of interface connected to an EBGP peer
Redistribution change
Reception of new paths for a route

RD (route distinguisher):

Make the tenant prefix unique in VPN environment

VPNv4 prefix: RD (64-bit) + IPv4 address (32-bit)

Default Route Advertisement:

network 0.0.0.0

RIB must have the default route

neighbor x.x.x.x default-originate

Only advertise the default route to the certain neighbor
Not necessary to have the default route in RIB

redistribute static

"default-information originate" must be configured
Not recommended

Route Redistribution:

By default, only EBGP routes will be redistributed into IGPs

Configure "bgp redistribute-internal" for IBGP routes

PAs of locally redistributed routes

Origin: ? (incomplete)
NH: the next-hop IP address in RIB
Weight: 32,768
MED: IGP metric

Considerations:

Scalability issue, for instance, router memory.
Loss of BGP path attributes
Route loops and suboptimal routes (e.g. route feedbacks and race conditions)

Route Summarization:

aggregate-address x.x.x.x x.x.x.x [option]

Installed in the BGP table only if a smaller route that is in the aggregate range is present
Create the null 0 summary route automatically
Aggregated route will lose PA information (such as AS-Path, MED and communities) due to aggregation (indicated by the atomic aggregate attribute)

Solution: add the option as-set

Option summary-only: suppress all smaller routes
Option suppress-map route-map-name: suppress the selective prefixes

Still possible to leak the route via "neighbor x.x.x.x unsuppress-map route-map-name"

Static way

Configure the static route "ip x.x.x.x x.x.x.x null0" and advertise it via the network statement
The summary route will always be advertised even if the network doesn't exist

~~auto-summary~~ (just let it go)

Route Reflector (RR):

Hub-and-spoke structure to scale IBGP sessions
Rules:

RR receives route updates from non-RR clients and it will send to RR clients only.
RR receives route updates from RR clients and it will send to RR clients and non-RR clients.
RR receives route updates from EBGP peers and it will send them to RR clients and non-RR clients
Note: the non-RR client can be the RR of other cluster.

Loop prevention:

ORIGINATOR_ID
CLUSTER_ID

Design considerations:

Redundancy: ideally, two or more RR per each service set (address-family) within AS
Hierarchical RR: multilayer RR structure
Partial RR:

Use BGP RR group
Use standard BGP community

Out-of-band RR: dedicated RR for route reflection and they are outside of the data path

Will not configure "next-hop-self" in it

搜索此博客

Qifawu

BGP Notes

评论

发表评论