In this doc, we will look at how we can, or prevent, summation of
groups of IPs within BGP. We will also look at how BGP uses metrics to
manipulate pathways from the end customer to te internet; with local
preferences and weights (bigger is better), and from the internet (same
AS#) to the end customer; with MED's and paths (smaller is better).
As mentioned in an earlier section, the number of routes that are advertised by a bgp router is set by each bgp network
statement. A poorly configured router that is, for example advertising
16 class C networks, could send out one route statement for each class
c network, and thus send lots of unnecessary traffic to other bgp
neighbors. This section will look at different ways to control the
amount of BGP advertisements sent, and ways to control them.
To lower the number of routes that need to be advertised, bgp
will allow us to group routes together so that we can advertise many
routes with just one route statement.
There are two main bgp commands that we can use to do this. The aggregate-address address mask [summary-only] and aggregate-address address mask [as-set] commands. The first combines a neighbors specific routes into one route that is then forwarded, and the second command
Examining how the Summary flag works, lets look at the example of the following three AS's.
Here, router B is sending out different route statements for
each one of its class C networks. Router A sees this, and rather then
passing all those separate route statements, it combines them into one
aggregate route (a /22 network). Thus we spare the next router from receiving extra unnecessary route statements.
| 1.1.1 aggregate-address summery-only command |
| Router A Config |
Router B Config |
1a01 ! -- aggregate neighbors routes
1a02 router bgp 100
1a03 aggregate-address 9.0.16.0 255.255.252.0 summary-only
1a04 network 5.5.1.0
1a05 neighbor 9.0.1.1 remote-as 150
1a06 neighbor 2.1.1.1 remote-as 50
1a07 ...
|
1b01 ! -- router with lots of advertised routes
1b02 router bgp 150
1b03 network 9.0.16.0
1b03 network 9.0.17.0
1b03 network 9.0.18.0
1b03 network 9.0.19.0
1b05 neighbor 5.5.1.1 remote-as 100
1b06 ...
|
You can view what received routes are aggregated by running a show ip bgp command.
include output from 'sh ip bgp' demonstrating this behavior
The
as-set attribute is useful if you are aggregating many different AS's
into one big route. This would be useful if you were a large isp who
dished out many contiguous class C networks for clients. Rather then
having to advertise each client (with a separate AS), you could
advertise one large network, and associate many AS's for it.
In this following example, router A is going to aggregate the
networks 50 and 100 into one route that it will then be passed to
router B.
| Aggregating routes with the as-set attribute |
1c01 ! -- combine AS's into one route
1c03 router bgp 150
1c03 aggregate-address 200.10.4.0 255.255.254.0 as-set
1c04 network 9.10.5.0
1c05 neighbor 200.10.4.1 remote-as 50
1c06 neighbor 200.10.5.1 remote-as 100
1c07 ...
|
Without aggregating, router A would send to router B two route statements like such:
200.10.4.0 100 50
200.10.5.0 100 150
But instead, with the as-set aggregation, Router B gets the following information:
200.10.4.0/23 100 [50 150]
The one caviot with this command is that Router B now, can not tell
how AS's 50 and 150 are arranged to Router A's AS 100. They could be one
behind the other, or side by side as in the picture below.
Originally we learned that to advertise a class C network we could use the network
command. If we wanted to advertise four consecutive class C networks
we would list four network statements in the BGP config. The results of
this would be four separate BGP statements that are advertised out to
that routers neighbor.
If you would rather send out one aggregate route for your four class C networks, you can do this by adding the mask flag to the network
statement. The following
| network command examples |
| Advertising four class C networks |
Advertising one /22 network |
1d01 ! -- four networks
1d02 bgp 10
1d03 network 195.10.16.0
1d04 network 195.10.17.0
1d05 network 195 10 18.0
1d06 network 195.10.19.0
|
1e01 ! -- one network
1e02 bgp 20
1e03 network 195.10.16.0 mask 255.255.252.0
1e04 ip route 195.10.16.0 mask 255.255.252.0 null 0
|
If you use the network command
twice, once with masking and then again without, the bgp router will
transmit both routes, even if the non masked route is a subset of the
masked one. This technique could be helpful if you had a large set of networks that you wanted to advertise, but that you also needed a few
parts of that network advertised on its own (maybe because it could be
advertised from somewhere else in case of a fail-over).
The network statement relies on the
fact that the network that it should advertise is also in the IGP
routing database. If part of the network that bgp should advertise is
missing, (ie a network flap), then the bgp routing table will also
flap. To avoid IGP creating bgp flaps, you can use floating static
routes and default metrics line in line 1e04.
Remember that BGP tells others how to route to you, but you can also use it to tell yourself how to route out to the internet. Some of the metrics below are for internal routing, or how you send traffic outside. Internal metrics would be "weight" and "local preference". There are other metrics which are used to suggest to other AS's on how they can send routes to you. External metrics would be MED's. The important thing to note about external metrics is that it is only a suggestion, because we are dealing with autonomous systems. If a carrier want's to strip your routing information, MED information, or aggregate you into other routes, that it their option.
- BGP Weight Metric
- BGP Local Preference
- BGP Multi Exit Discriminator
- How BGP Selects a Path
If you have a router that has multiple neighbors you can have that
router "weight" incoming routes better from one neighbor over the other
using the neighbor {ip-address|peer-group-name} weight {weight}
command. These incoming weights are specific to that router, as it
does not pass the preference on to any of it's peers. The default
weight is set at 32,768, and the higher weighted route will be preferred.
For example, this can become very useful if you have a router
with multiple BGP neighbors. For example, if you had a router that had
two BGP peers, one connected via a DS3 link, and a backup connected via
a T1. You would want to favor the DS3 link more, and thus weight it
higher then the T1.
| Weight Attribute |
1f01 ! -- local weights, higher better
1f02 bgp 10
1f03 neighbor 200.10.50.1 remote-as 200
1f04 neighbor 200.10.50.1 weight 50000
1f05 neighbor 400.40.50.1 remote-as 400
1f06 neighbor 400.40.50.1 weight 40000
|
Note that since this only affects the routing table on that specific
router, it is a great way of controlling outbound traffic if the router
has more then one interface, but (since the weight is not advertised)
it is not very helpful if there are two routers, each with one
interface.
The
previous command was useful for giving preferences to routes on a
single router. This command gives preference to routing information
ibgp wide. It is used within an AS to find the best route to leave the
AS. (in other words, when you get routes from other carriers, you can tell traffic to prefer router "A" over "B", but this routing information is local to you, and not sent out to the carriers.)
The command bgp default local-preference {value} specifies different ibgp routers with different weights. These weights
are attached onto the ibgp route information as it's passed on to all
local, ibgp routers. The default weight value is 100, the higher number is preferred, and the range is 0 - 4,294,967,295 = ( 2^32).
In the example below, the routers in AS10 are all ibgp neighbors. Of
its two egress routers (A & B), B's link out is a smaller pipe, a
T1, vs. A's link which is a DS3.
To do this, we can use the bgp default local-preference {value}
command in both the A and B routers. By specifying A's Local Preference
as a higher value, we can pass to all the other ibgp routers the fact
to prefer the A router when sending outbound traffic.
We want to tell all the routers within AS10 to prefer sending
outbound traffic through the A router. In this example, when traffic
should be sent to the AS 400, the ibgp route table will say that it has
the same number of hops to get there. But that traffic to router A is preferred. So the ibgp routers will send traffic to the A router, and
then through AS 200 to get traffic to AS 400.
| BGP Local Pref |
| Router A (prefered) |
Router B |
1g01 ! -- preferred, faster, outbound route.
1g02 router bgp 10
1g03 neighbor 5.10.10.1 remote-as 10
1g04 neighbor 193.200.200.1 remote-as 200
1g05 bgp default local-preference 400
|
1h01 ! -- backup, slower, outbound route.
1h02 router bgp 10
1h03 neighbor 5.10.10.2 remote-as 10
1h04 neighbor 209.300.300.1 remote-as 300
1h05 bgp default local-preference 110
|
BGP Local Pref modifies how traffic leaves your AS. MED's modify how traffic arrives, but only from a single peer who you might have two links with. (Maybe a DS3, and a backup T1 or something.)
When
two AS's have multiple links with each other, one AS can tell the other
which peer point it would prefer traffic to enter through via the bgp med value command. This value can be utilized by the neighboring peer, but will not be propagated beyond. (IE: MED is an optional non-transitive attribute.)
add more notes and example for MED's
The lower MED value is preferred over the higher one.
| BGP MED |
| Router A (prefered) |
Router B |
1m01 ! -- preferred, faster, outbound route.
1m02 router bgp 10
1m03 neighbor 5.10.10.1 remote-as 10
1m04 neighbor 193.200.200.1 remote-as 80
1m05 neighbor 193.200.200.1 route-map medout out
1m06 !
1m07 route-map medout permit 10
1m08 set metric 0
|
1n01 ! -- backup, slower, outbound route.
1n02 router bgp 10
1n03 neighbor 5.10.10.2 remote-as 10
1n04 neighbor 209.300.300.1 remote-as 300
1n04 neighbor 209.300.300.1 route-map medout out
1n04 !
1m07 route-map medout permit 10
1m08 set metric 20
|
to enter through via the bgp med value comm
Still need to talk about prepending (aka: route poisoning)
router bgp 10
neighbor 192.168.1.1 remote-as 100
neighbor 192.168.1.5 remote-as 100
neighbor 192.168.1.5 route-map PREPEND out
network 10.1.1.0 mask 255.255.255.0
!
route-map PREPEND permit 10
set as-path prepend 200 200 200
! |
- if NextHop is unreachable, do not use that update
- Prefer the path with the largest Weight
- If no Weight or the same Weight, select the largest Local Preference.
- If same Local Preference, prefer the path (if any) that was originated by BGP on this router.
- If no route was originated, prefer the shorter AS Path.
- If all paths are the same length, prefer the lowest origin code: ( IGP < EGP < Incomplete )
- If origin codes are the same, prefer the path with the lowest MED.
- If path are the same MED, prefer the External path over Internal.
- If path are still equal, prefer the path through the closest IGP neighbor.
- Prefer the path with the lowest BGP router ID.
If a given network is learned by a different routing protocol it is assigned a default administrative distance as it is interpreted into the
current routing protocol. Of all the possible routing protocols that
the route is learned from, the one with the lowest administrative
distance is the weight assigned.
If administrative distance is equal because a route is learned
from multiple neighbors, but all neighbors use the same routing
protocol, then take the lowest metric to select the best route for a
given network.
| Admin Distances for Route Selections |
| Directly Connected Interface |
0 |
| Static Route - (interface specified) |
0 |
| Static Route - (next hop address specified) |
1 |
| Enhanced IGRP Summary Route |
5 |
| External Enhanced IGRP |
20 |
| IGRP |
90 |
| OSPF |
100 |
| IS-IS |
115 |
| RIP |
120 |
| EGP |
140 |
| Internal BGP |
200 |
| Unknown |
255 |
Administrative distances are set by cisco routers by default to
the above values, but these values are adjustable. For example, the
following command could be used to change the administrative distance
for routes being learned from RIP.
| Changeing the Admin. Distance for RIP |
1i01 ! -- Change RIP Admin Distance from default "120" to "99"
1i02 router rip
1i03 distance 99
|
All routers in an Autonomous System (AS) should use the same
administrative distance matrix. If you change it in one place, you
should change it on every router to allow for consistent routing across
your AS. Once you change the administrative distance its new value will
be automatically applied to all incoming IP routing updates.
As
noted above, the BGP tie-breaker for choosing a pathway is to take the
lowest router ID when everything else is equal. To allow for load
balancing, the loopback address is used as the next-hop address between
the two routers, for all routes out of the links to be load balanced.
Since BGP will have multiple equal paths between the two loopback
addresses, load balancing will occur.
To make this work we use two different commands, the neighbor {} ebgp-multihop, and the neighbor {} update-source commands.
The first, the neighbor {ip-address | peer-group} ebgp-multihop
command, allows an external BGP neighbor to not be directly connected,
but instead, allows the two to be separated by a few routers. Because
of this, the remote router must be in the routers IP routing table, so
there are should be static routes for that route.
The neighbor {ip-address | peer-group} update-source [interface-type] [interface-number]
command allows a router to use a specified interface as the update
source and advertise that interface to other internal BGP routers. Generally, this command is used to specify the loopback interface as the
update source. You should prefer the BGP interfaces if you are load
balancing, because otherwise, if the one interface that is used as the
peer goes down, then all the routes are considered not valid, even if
there are many other load balanced interfaces that are still up.
The following is an example configuration between two load balanced routers.
| Example of load balancing |
| Router A |
Router B |
1j01 ! -- LB Interfaces
1j02 interface loopback0
1j03 ip address 10.100.1.1 255.255.255.255
1j04 router bgp 100
1j05 network 190.100.0.0 mask 255.255.252.0
1j06 network 190.100.100.0 mask 255.255.255.252
1j07 network 190.100.100.4 mask 255.255.255.252
1j08 neighbor 10.100.2.1 remote-as 200
1j09 neighbor 10.100.2.1 update-source loopback 0
1j10 neighbor 10.100.2.1 ebgp-multihop
1j11 ip route 10.100.2.1 255.255.255.255 190.100.100.2
1j12 ip route 10.100.2.1 255.255.255.255 190.100.100.6
|
1k01 ! -- LB Interfaces
1k02 interface loopback0
1k03 ip address 10.100.2.1 255.255.255.255
1k04 router bgp 200
1k05 network 57.200.200.0 mask 255.255.254.0
1k06 network 190.100.100.0 mask 255.255.255.252
1k07 network 190.100.100.4 mask 255.255.255.252
1k08 neighbor 10.100.1.1 remote-as 100
1k09 neighbor 10.100.1.1 update-source loopback 0
1k10 neighbor 10.100.1.1 ebgp-multihop
1k11 ip route 10.100.1.1 255.255.255.255 190.100.100.1
1k12 ip route 10.100.1.1 255.255.255.255 190.100.100.5
|
