Bigbgp (1)

Intro to BGP
All-Day
Tutorial

Avi Freedman
freedman@netaxs.com

Index
• Internet Connectivity Overview
• Multihoming Concepts
• Multihoming Without BGP
• Multihoming - Address Space Complications

Index
• Basic BGP - The BGP Route
• Basic BGP - Inserting Routes into BGP
• Basic BGP - Advertising Routes
• Basic BGP - Other BGP Route Attributes
• Basic BGP - Selecting Routes

Index
• Multihoming with BGP - an Introduction
• Interlude - Hardware for BGP
• Multihoming with BGP with a Cheap Router
• Multihoming with BGP - Taking Just
Customer Routes
• Multihoming with BGP - Taking Full Routes
• Default Routing in BGP

Internet
Connectivity
Overview

Having Internet Connectivity
• To have complete Internet connectivity you
must be able to reach all destinations on the
net.
• Your packets have to get delivered to every
destination. This is easy (default routes).
• Packets from everywhere else have to “find
you”. This is done by having your ISP(s)
advertise routes for you.

Multihoming Without BGP
• To get Internet connectivity, you can just
default route your traffic to your upstream
providers.
• To get traffic back from the Internet, you
need to have your providers tell all of the
rest of the Internet “where you are”.

BGP Route Advertisement (1)
• Think of a BGP route as a “promise”.
• If I advertise 207.8.128.0/17, I promise that
if you deliver traffic to me for anywhere in
207.8.128.0/17, I know how to deliver it at
least as well as anyone else.
• If my customer has 207.8.140.0/24, I
generally will not announce that route
separately since it is covered by my
207.8.128.0/17 aggregate route.

• By making sure these routes, or “promises”,
are heard by ALL providers on the ‘net,
your provider ensures a return path for all of
your packets.
• Remember, sending packets OUT is easier
than getting them back.
• Also, remember - sending routes OUT
causes IP traffic to come IN.

• But the most specific route wins, so if one
of my customers’ ISPs is advertising
207.8.240.0/24, all incoming traffic from
other networks will start flowing in that
pipe.
• So I must “punch a hole” in my aggregate
announcement and advertise 207.8.128.0/17
and 207.8.240.0/24.

• The complete set of routes advertised by all
BGP speakers on the net is about 55,000
routes as of 10/98.
• If your route is missing in the “view” of any
major provider, you will not have
connectivity to them.

Multihoming
Without BGP -
How it Works

Customer Side - Outbound
• All you need to do is to put in static default
route(s). To prefer two upstreams equally:
– ip route 0.0.0.0 0.0.0.0 s4/0
– ip route 0.0.0.0 0.0.0.0 s4/1
• To use one link as a backup only for
outbound packtes:
– ip route 0.0.0.0 0.0.0.0 s4/0
– ip route 0.0.0.0 0.0.0.0 s4/1 10
– why? S4/1 could be a 56k or backup link

Cisco Load Balancing
• The way Ciscos (except for big new ones
running “CEF”) work if there are two
“equal-cost” routes to the same place is -
– Option 1 - Round-robin the packets without
“route caching”. This goes through the slowest
sections of the router’s OS. Bad. Also, if you
are connected to different ISPs, packets can
arrive out of order, etc…
– Option 2 - Use route caching (default). Traffic
to the same dest IP will always use the same
interface, until the cache entry expires.

Customer Side - Inbound
• Just tell your ISP what address space you
are bringing, if any.
• Your ISP may allocate you space out of
their larger address blocks.
• If so, they need to announce your space
“more specifically”.
• But you do no work other than tell your ISP
what to do.

Provider Side (1)
• If both providers don’t advertise your routes
with the same specificity, you might have -
– netaxs saying “4969 sez 207.8.128.0/17”
– uunet saying “701 sez 207.8.195.0/24”
• Bad, because almost all traffic on the ‘net
will come into you via UUNET.
• {note} - talk about address filters

Provider Side (2)
• What you need is -
– uunet saying “701 sez 207.8.195.0/24”
• Good, because -
– 1) Because the two 207.8.195.0/24 routes are of
the same specificity, providers CAN choose
btwn netaxs and uunet to get to you; and
– 2) For some people who don’t listen to /24s and
such in new address space, they still have the
207.8.128.0/17 route to use to get to you.

Address Space Complications
• So, in the case of -
– uunet saying “701 sez 207.8.195.0/24”
• “Some people won’t listen to the /24, so
what happens if my netaxs connection goes
down?”
• Not a problem!!! Because netaxs will hear
the UUNET /24. Sprint send traffic to
netaxs; netaxs to uunet; and uunet to you.

Disadvantages of
not using BGP
• You gain a bit more control of your destiny
when you speak BGP yourself. You can
break up your routes in an emergency, or to
tune traffic. You can “pad” your
announcements to de-prefer one or more
upstreams.
• Also, you lose the ability to fine-tune
outbound traffic flow to the “best”
upstream.

Why BGP?
• BGP is a multi-vendor “open” protocol with
multiple implementations, all mostly
interoperable. It is the only actively used
EGP on the Internet.
• The main design feature of BGP was to
allow ISPs to richly express their routing
policy, both in selecting outbound paths and
in announcing internal routes. Keep this in
mind as we progress.

BGP is … (1)
• An Exterior Gateway Protocol (EGP), used
to propagate tens or hundreds of thousands
of routes between networks (ASs).

• The only protocol used to do this on the
Internet today.

BGP is … (2)
• The Border Gateway Protocol, currently
Version 4 - defined in RFC 1771, and
extended (with additional optional
attributes) in other RFCs.
• A “distance-vector” routing protocol,
running over TCP port 179.
• Supports modern “classless” routing.
BGP3, RIPv1, and some others do NOT.

Purpose of BGP
• To allow networks to tell other networks
about routes (parts of the IP address space)
that they are “responsible” for.
• Using “route advertisements”, or
“promises” - also called “NLRI” or
“network-layer reachability information”.
• Networks are “Autonomous Systems”.
• Identified in BGP by a number, called the
ASN (“Autonomous System Number”)

Basic BGP Concepts (1)
• BGP exchanges routes between ASs.
• When routes are exchanged, ASNs are
stamped on the routes *on the way out* -
adding one “AS hop” per network
traversed. (0-65535)
• No concept of pipe size, internal router hop-
count, congestion - in some sense BGP
treats all ASs the same.
• ASs allow administrative debugging,
“policy” routing, and *loop detection*.

BGP AND ASNs
AS 12001

AS 4969
AS 6461

AS 701 AS 5000

• Routes are exchanged over “peering
sessions”, which run on top of TCP.
• Keepalives are used to avoid needed to re-
send the whole table periodically.
• The routes are “objects”, or “bags” of
“attributes” - really mini-databases.
• BGP is actually two protocols - iBGP,
designed for internal routing, and eBGP,
designed for external routing.

• There is only one “best” BGP route for any
given IP block at one time.
• This “best” BGP route is not always the
route that gets “installed” into the router’s
RIB/FIB.
• Once a session comes up, all best-routes are
exchanged. Then over time, just “topology
updates” are exchanged.
• You can ONLY exchange “best” routes.

• Policy
– The Internet was a strange place before the
modern commercial Internet evolved in 1992-
1993.
– Some networks had policies about what kind of
traffic they would carry.
– BGP was designed to allow network operators
to make routing decisions based on whatever
“policy” they wanted (or HAD) to use.

CISCO DIAGRAM -
RIB
FIB
ETC

Basic BGP Concepts -
The BGP Route
and
Route Attributes

The BGP Route
• A BGP “route” is a “bag” of objects, or
“attributes”.
• The “prefix” is the section of address space
being advertised. A prefix consists of:
– A starting point (i.e. 207.8.128.0)
– A netmask (i.e. /24, aka 255.255.255.0)

What Is an Attribute?

... Next AS ... ...
MED
Hop Path

• A BGP message consists of a prefix and information about
that prefix (i.e., local-pref, med, next-hop, originator, etc...).
Each piece of information is encoded as an attribute in a TLV
(type-length-value) format. The attribute length is 4 bytes
long, and new attributes can be added by simply appending a
new attribute.
• Attributes can be transitive or non-transitive, some are
mandatory.

Next Hop Attribute

• Next-hop IP address to AS 6201
reach a network. 198.3.97.0/24
A
• Router A will advertise .202
207.240.24.200/30
198.3.97.0/24 to router B with a next-
hop of 207.240.24.202.
.201 AS 3847
• With IBGP, the next-hop does not
change. B

• IGPs should carry route to next-hops,
using intelligent forwarding decision.
C

Next Hop Self

198.32.184.116
198.32.184.42
AS701 A B AS3561

D
AS3847 C AS1
198.32.184.56 198.32.184.19

AS Path Attribute (1)
• Sequence of AS(s) a route has
AS3561
traversed. 204.70.0.0/15 AS701
• Provides a mechanism for loop 192.67.95.0/24
G
detection. F
• Policies may be applied
based on AS path. D
AS3847
• Local AS added only when 207.240.0.0/16 C
AS1673
140.222.0.0/16
send to external peer. B
* Shortest AS path preferred
E

A
192.67.95.0/24 3847 701 i
140.222.0.0 3847 1673 i
AS6201 204.70.0.0/15 3847 3561 i
207.240.0.0/16 3847 i


• Sprint is 1239; UUNET is 701; Net Access
is 4969.
• When pattern-matching, or regexping,
AS_PATHS, ^ means “match beginning”,
and $ means “match end”.
• The null AS-Path is ^$ - if the AS-Path is
null, the BGP route originated inside the
same AS.


• ^1239 4969$ is how a Sprint customer
would see a Net Access route.
• ^1239 4969 11023$ is how a Sprint
customer would see a Net Access BGP
customer’s route.
• ^4969 11023$ is how Sprint itself sees that
same route.

Multi-Exit Discriminator (MED)
(MED

• Indication to external peers of the preferred
path into an AS.
• Affects routes with same AS path.
• Advertised to external neighbors
• Usually based on IGP metric
* Lowest MED preferred

MED Attribute (2)
• The MED (multi-exit discriminator) is a
commonly used attribute. It comes after the
AS_PATH in evaluation, and thus isn’t
quite as much of a “hammer” as local-pref.
• Commonly, MED is used to tack a distance
on BGP routes as they move within your
network.
• NSPs advertise MEDs to each other to let it
be known which POP the route is “closest”
to.

MED Attribute (3)

+5
+40

+20

• Applies on a AS path basis
• Current aggregation schemes significantly lessen value.

Origin Attribute
• One of the mandatory, but minor, attributes
of a BGP route is the origin. It is one of (in
order of preference):
– IGP (i) (from a network statement)
– EGP (e) (from an external peer)
– Unknown (?) (from IGP redistribution)
• It can be re-set, but that is not often done.
• It is almost-last in the selection algorithm.

Weight Attribute

• Cisco proprietary, not part of any spec.
• Local to router.
• Value 0-65535 (default if originated by
router - 32768, other - 0)

* Highest weight preferred

Weight Attribute (ctd)
• Weight is rarely used. It overrides almost
all other attributes in the decision path, and
is local to a specific router - it is never sent
to other routers, even ones inside your
ASN.
• Usually used for temporary “I-don’t-have-
time-to-think-about-it” fixes.

Local Preference Attribute
AS 3847
F E
G

C D

208.1.1.0/24 80 208.1.1.0/24 100
Preferred by all
• Local to AS AS3847 routers
A B
• Used to influence BGP 208.1.1.0/24

path selection AS 6201
• Default 100
* Highest local-pref preferred

Local-Pref Attribute (2)
• An often-used attribute, local-pref
(normally 100) overrides AS_PATH, and is
transitive throughout your network. It is
never advertised to an eBGP peer.
• For example, you can express the policy
“prefer private interconnects” by making
the local_pref be 150 and leaving all other
peers at 100.
• Best used as an intermediate-level knob.

iBGP vs. eBGP
• BGP is very strange. It is promiscuous with
external routes, making it very easy for you
to become “MAE-Clueless”, yet it makes it
very hard to advertise routes thoroughly
inside your network.
• iBGP sessions are established when peering
with the same AS; eBGP otherwise.
• Same protocols; different route install rules.
• YOU MUST STRONGLY FILTER ALL
eBGP SESSIONS!

iBGP
AS 3847
When BGP speakers in the same
AS form a BGP connection for A
the purpose of exchanging routing
information, they are said to be
running IBGP or internal BGP. c
B

IBGP speakers are usually
fully-meshed.

eBGP (1)
AS 3561

A
When BGP speakers in different
ASs form a BGP connection for
the purpose of exchanging routing
AS 3847
information, they are said to be
running EBGP or external BGP. B

EBGP peers are usually directly
connected.

eBGP (2)

AS 2033
AS 7007

AS 4200
AS 2041

iBGP and eBGP Diagram

AS 1239

XP AS 701
AS 7007

AS 6079
AS 4006

eBGP Rules
• By default, only talks to directly-connected
router.
• Sends the one best BGP route for each
destination.
• Sends all of the important “attributes”;
omits the “local preference” attribute.
• Adds (prepends) the speaker’s ASN to the
“as-path” attribute.
• Usually rewrites the “next-hop” attribute.

iBGP Rules
• Can talk to routers many hops away by
default.
• Can only send routes it “injects”, or routes
heard DIRECTLY from an external peer.
• Thus, requires a FULL mesh.
• Sends all attributes.
• Leaves the as-path attribute alone.
• Doesn’t touch the “next hop” attribute.

Logical view of 16 routers, fully
meshed

iBGP Restriction (1)

Assume AS1239 sends route 10.0.0.0/8 to
AS2828. Router A will send that route to
Routers B and C.

B

AS 2828
C A AS 1239

iBGP Restriction (2)
When Router B receives 10.0.0.0/8, it will not
propagate that route to Router C because it
was learned from an iBGP neighbor.
Router C will behave similarly.

B

AS 2828
C A AS 1239

iBGP and next-hop (1)
Furthermore, the Next Hop for 10.0.0.0/8 will
be the serial interface on the AS1239 router,
even in Router B’s and Router C’s
forwarding table.

B

AS 2828
C A AS 1239

iBGP and next-hop (2)
• With iBGP, next-hop is not a router directly
connected.
• So a “recursive lookup” is needed.
• After the next-hop is found, a second
lookup is made to figure out how to send
the packet “in the direction” of the next-
hop.

Basic BGP Concepts
Inserting Routes
into BGP

Inserting Routes into BGP (1)
• How do routes get into BGP? They have to
come from somewhere. You have to insert
routes into BGP, and someone had to insert
external routes that you get into BGP
somewhere else in the first place.
• Two main ways:
– network statements (like static BGP routes)
– redistributing from OSPF, static, etc...

• network statements
– “network x.y.z.q [mask a.b.c.d]”
– MUST have an EXACTLY-matching IGP route
- specificity must be an exact match
– Doesn’t scale beyond 200 or so network
statements per routers; not a problem, though.
– Makes scaling easier when you have to support
multi-homed customers

• aggregate-address statements
– “aggregate-address x.y.z.q a.b.c.d [aggregate-
only] [suppress-map XXX]”
– (Really a relative of the network statement)
– Brings up the given network if there are any
more specific BGP routes for the prefix
specified.
– Usually used with aggregate-only to suppress
more specifics.
– Usually used in conjunction with redistribution.

• Redistribution
– ALWAYS redistribute through an address
filter! Otherwise you will have crud in your
BGP!
– Examples later on...
• Default route is a special case. More soon.

BGP Peering Sessions (1)
• BGP Routes are exchanged inside of BGP
peering sessions.
• BGP uses TCP to ensure reliable delivery of
routing updates.
• If a TCP session dies, all associated routes
must be withdrawn.
• BGP peers, or neighbors, must be specified
explicitly. This is a good thing.

BGP Peering Sessions (2)
• Once a peering session is set up:
– Both sides flood the other end with all of their
best BGP routes. VERY IMPORTANT - there
is one best route per prefix, and that is the route
that is advertised. BGP can only advertise
routes that are eligible for use or routing loops
can occur.
– Then, periodic updates send new routes and/or
withdraw old ones, and keepalives are sent
every N seconds.
– On a very stable network,very little or no traffic
should flow besides keepalives.

Peering - BGP State Machine
• There is a state machine that describes the
setting up, use, and tearing down of BGP
sessions. It’s useful to know the states
because Cisco uses them to describe session
state.
• Idle -> Connect -> Active {send “startup”
packet} -> OpenSent -> OpenConfirm
{wait for ack} -> Established [ … -> Idle]
• In “sho ip bgp summ”, “Active” does NOT
mean Active, it means “waiting” - FYI.

Peering - Processing Routes
• For each route received:
– If it’s a valid route AND passes any filters, it
must be put into the BGP routing table.
– Then, unless it is replacing a duplicate, a best-
path computation must be run on all candidate
BGP routes of the same prefix.
– Then, if the best route changed, the RIB and/or
FIB must be updated.
– This process is done for ALL incoming BGP
routes.

Filtering BGP Routes -
BGP Policy Control

BGP Policy Control
• To decide what routes can and can’t go to
various other routers, you can “filter” using:
– “distribute lists” (“prefix filters”) - lists of
routes
– “filter lists” (“as-path filters”) - lists of regular
expressions matching or denying ASs
– “route maps” (“BGP Basic programs”) that
allow you to match and change most BGP
attributes

Distribute List (1)

• Per neighbor access list applied to BGP routes
• Inbound or outbound
• Based upon network numbers

Distribute List (2)
router bgp 3847
neighbor 207.240.8.246 remote-as 8130
neighbor 207.240.8.246 distribute-list 127 in
neighbor 207.240.8.246 distribute-list 101 out

access-list 127 permit ip host 207.19.74.0 host 255.255.255.0
access-list 127 deny ip any any
! explicit deny if not specified

Distribute List (3)
access-list 10 deny ip 10.0.0.0 0.255.255.255
access-list 10 permit ip any

A sanity filter like this keeps your table neat and prevents you from advertising
crud to your peers.

Filter List (1)

• Filter routes both inbound and outbound based on value of AS
path attribute.
• Called “as-path” access, or filter, lists.
• Configuration
router bgp 3847
neighbor 207.240.10.100 remote-as 2900
neighbor 207.240.10.100 distribute-list 100 in
neighbor 207.240.10.100 distribute-list 101 out
neighbor 207.240.10.100 filter-list 10 in

ip as-path access-list 10 permit ^2900$
ip as-path access-list 10 deny .*

Cisco Regular Expressions (1)

. Period matches any single character,
including white space.
* Asterisk matches 0 or more sequences of
the pattern.
+ Plus sign matches 1 or more sequences of
the pattern.
? Question mark matches 0 or 1 occurrences
of the pattern


^ Caret matches the beginning of the input
string.
$ Dollar sign matches the end of the input
string.
_ Underscore matches a comma (,), left brace
({), right brace (}) left parenthesis, right
parenthesis, the beginning or end of the
input string, or a space.


[] Square brackets designate a range of single
character patterns.
- Hyphen separates the endpoints of a range.

As you may have noticed, these are much like
standard vi regular expressions.

Applying AS Path Filtering
6201 701
A
E F
B
3847

6202
C D
G

The following configuration could be used on router B to
accept routes from AS6201 & 6202 and deny all others.

ip as-path access-list 10 permit ^6201$
ip as-path access-list 10 permit ^6201_6202$

netaxs AS-Path ACLs

• 3 default lists
• (Permit all; Deny all; Permit only our
routes)

ip as-path access-list 1 permit .*
ip as-path access-list 3 permit ^$

Route Maps (1)

Route-maps are cisco’s mechanism to select
and modify routes with if/then style
algorithms.
Route-maps are used for more than just BGP
in a cisco router, such as traffic shaping and
policy routing.

Route Maps (2)

Route-maps follow this format:

route-map <name> <per|deny> <#>
[match statements]
[set satements]

[repeat with unique sequence
numbers as needed]

Route Maps (3)

Route-maps follow this format:

route-map <name> <per|deny> <#>
[match statements]
[set satements]

[repeat with unique sequence
numbers as needed]

Route Maps (4)

For route-maps with the keyword “permit”, if
the prefix being examined passes the match
statement, the set commands are executed
and the route-map is exited.
If the match statement is not passed, the next
sequence number is executed.
If there are no more sequence numbers, the
prefix is filtered/dropped.

Route Maps (5)

For route-maps with the keyword “deny”, if
the prefix being examined passes the match
statement, the prefix in question is filtered
and no more sequence numbers are
executed.
If the prefix does not pass the match
statements, the next sequence number is
executed.

Selecting BGP Routes

• Usually there will be 2, 3, 4, etc… ways to
get to a given destination, all of which are
represented by BGP routes.
• There is a way of picking the “best” one.
• Most important note -
– Selection is NOT random between “similar”
routes.
– You can ALWAYS figure out why something
is happening if you understand the rules.

Selecting BGP Routes - Basic
• ALWAYS find the most specific route.
• ONLY consider paths w/ reachable
NEXT_HOPs.
• Prefer a route originated on the local rtr.
• Then, unless tuning has been done, pick the
route with the shortest AS-PATH; then
origin code; select on MED; then router ID.
• Or, if weight, LOCAL_PREF is set, or
padding done to AS_PATH, look at those.

BGP Decision Algorithm
• Do not consider IBGP path if not synchronized
• Do not consider path if no route to next hop
• Highest weight (local to router)
• Highest local preference (global within AS)
• Prefer local route
• Shortest AS path
• Lowest origin code IGP < EGP < incomplete
• Lowest MED
• Prefer EBGP path over IBGP path
• Path with shortest next-hop metric wins
• Lowest router-id

Router Vendors (1)

• cisco (affectionately called Crisco)
• Bay Networks (called Bad Networks)
• Livingston (called Lucington)
• Ascend (pronounced ASS-END)
• PC router w/ unix and gated

Router Vendors (2)

• Cisco rules the backbone router market.
Everyone runs Cisco code, so everyone hits
the same bugs.
• Bay is next-biggest, but not really catching
up. Have their own working BGP
implementation, but takes them a while to
add new features. Major disadvantage -
GUI interface. Command-line is coming
but not yet.

Router Vendors (3)

• Livingston has a BGP implementation, and
though it isn’t feature-ful, it basically works
and is efficient (but only supports t1-speed
routing).
• Ascend runs gated and is nasty. They
barely understand bridging. You have to
reboot their routers to reload a config.

Router Vendors (4)

• The gated consortium, run by merit, puts
out a program capable of running OSPF and
BGP (among other protocols). Put on a
Unix box, it turns it into a router of sorts.
– Disadvantages -
• spinning hard disk
• “bgp assembly language”
• poor documentation/support

Cisco Products
• Can’t run full BGP -
– 2501 - 68030, 25mhz, 16mb ram max
– 4000/4000M/4500, 16-32mb max
• Can run full BGP -
– 2600 - risc,64mb max - < 3k
– 3620/3640 - risc, 64/128mb max 3-10k
– 4500M/4700 - risc, 64/128mb max 5-10k
– 7206 - risc, 128mb max 12-30k
– 7000 - motorola, 64mb max 4-30k
– 7500 - risc, dist switch, 128mb max 15-70k
– GSR/BFR - “anything you want” 30-120k

Multihoming with BGP
An Introduction

Step 1 - Determine Policy

• “You go find out what they want; we’ll start
programming the routers” doesn’t work
well.
• Before you step up to the router, determine
what routing policy you want to express
with your configuration.
• Plan your configuration, and ask how it
could put you (in an unwelcome light) on
the nanog mailing list.

Policy for Basic Multi-Homing

• We want to advertise our routes - all of
them, but only OUR routes. So, assemble a
list of our routes and masks.
• We want to accept all routes and let the
router sort them out, initially based on AS-
PATH length. If we don’t have enough
memory to take full routes, we’ll start off
taking none and then play later.

Warning - I am Blackholio (1)

• Never blackhole someone.
• Say www.uu.net is 137.239.5.24, and the
best match for that IP is the prefix
137.239.0.0/16.
• What happens if you announce
137.239.5.0/24, by accident or on purpose?
• Worldcom’s lawyers show up at your doors
and you look like an idiot.


• What happens if you have a T1 to Sprint
and a T1 to UUNET, and you announce
Sprint routes to UUNET? (Assume no
sanity filters at the upstream, which is
always a good assumption).
• Answer - you have become MAE-Clueless,
and all of UUNET tries to get to Sprint
through your T1.
• Why?


• As your provider, I have to believe that
your route is the best way to get to a given
prefix.
• Why? Because otherwise I can’t transit you
- I can only send routes to the other
providers on the Internet if I believe they
are the best ones.

Multihoming -
Minimal BGP

(for cheap routers)

Insert Static Default Routes
• Insert static default routes, either load-
balanced or with primary/backup, as per
non-BGP multihoming.
• Either
ip route 0.0.0.0 0.0.0.0 s4/0
ip route 0.0.0.0 0.0.0.0 s4/1
• Or
ip route 0.0.0.0 0.0.0.0 s4/0
ip route 0.0.0.0 0.0.0.0 s4/1 250

Gather Networks
• Routes
– 207.8.200.0/22
– 198.69.44.0/24
• Holdup routes keep the routes in BGP so they
don’t “flap”. “Flapping” can blackhole you.
• Then, build access-list and holdup routes
access 55 permit 207.8.200.0 0.0.3.255
access 55 permit 198.69.44.0 0.0.0.255
ip route 207.8.200.0 255.255.252.0 null0 250
ip route 198.69.44.0 255.255.255.0 null0 250

Set up BGP Base Config

ip as access 1 permit .*
ip as access 2 deny .*
ip as access 3 permit ^$

router bgp 22222
no sync
net 207.8.200.0 mask 255.255.252.0
net 198.69.44.0 mask 255.255.255.0

Configuring Neighbors - Note

• The best way to configure a neighbor is to
use cut-and-paste, or to tftpboot a snippet or
whole config.
• You have 30-60 seconds to type in the
whole neighbor clause before the session
could come up and start receiving and
sending routes - WITHOUT FILTERS if
you didn’t type fast enough...

Neighbor Configuration (1)

router bgp 22222
neigh 207.106.2.45 descr transit to netaxs
neigh 207.106.2.45 remote-as 4969
neigh 207.106.2.45 next-hop-self
neigh 207.106.2.45 version 4
neigh 207.106.2.45 dist 55 out
neigh 207.106.2.45 filter 3 out
neigh 207.106.2.45 filter 2 in

Neighbor Configuration (2)

router bgp 22222
neigh 10.40.4.81 descr transit to UUNET
neigh 10.40.4.81 remote-as 701
neigh 10.40.4.81 dist 55 out

Test it

• Do a “sho ip bgp”. Only your 2 routes
should show.
• Do a “show ip bgp neigh <neighip> adv”.
You should show that you are advertising
those 2 routes to your 2 neighbors.
• Go to nitrous.digex.net or another BGP
looking glass, to see that the routes are
being advertised under your AS, not the
provider’s, and that both paths are there.

Multihoming with BGP -
Taking Customer Routes

(an intermediate solution)

Taking Just Customer Routes

• One option in-between default routing and
taking full BGP is to at least take customer
routes from each provider.
• This way, you’ll be able to make some
intelligent decisions, which can be
especially important for news feeding and
dns and mail exchange optimization.
• If your provider isn’t Sprint or CW, you can
probably fit “customer” routes in 16mb.

Taking Just Customer Routes (2)

• The best plan is to get your provider to
advertise their customer routes ONLY to
you. Still, use the KGB motto - “Trust, but
verify”.
• Doesn’t work on small routers if your
upstream is MCI or UU.
• Or, community-based filtering (more later).


• So, a sanity filter:
ip as acc 10 deny _701_
ip as acc 10 permit .*
• (Prevent hearing routes from the big boys -
eve)

router bgp 22222
neigh 207.106.2.45 descr transit to netaxs
neigh 207.106.2.45 remote-as 4969
neigh 207.106.2.45 distribute 55 out

Multihoming with BGP -
Taking Full Routes

Policy
• Actually, very easy.
• Continue to advertise your routes, as before.
• Take full routing info.
• Later on, you can tune if you find that as-
path is not a good indicator to some sites.

So, what Policy?
• We’ll do the same thing on advertisement,
but we’ll take all routes from both
upstreams.

Configuring Full BGP
• Router bgp 22222
– neigh 207.106.2.45 remote-as 4969
– neigh 207.106.2.45 next-hop-self
– neigh 207.106.2.45 version 4
– neigh 207.106.2.45 distribute 55 out
– neigh 207.106.2.45 filter 3 out
– neigh 207.106.2.45 filter 1 in

Logistics of
becoming
Multihomed

Multihoming Logistics
• Address space.
• Redundant connectivity during switch.
• Test configs.
• Bring up outbound BGP first.

Multihoming to the
same Provider

Multihoming to the same Provider
• Same configs as being multi-homed, except
that as-paths will be the same from your
upstream, and your routes will be the same
in their network.
• Need some way of preferring. Either:
– Send more specifics and MEDs, and take
MEDs
– Play w/ route-maps and set some set of routes
higher through one POP than another

The Network Grows -
Supporting BGP Customers

Supporting BGP Customers (1)
• So you’ve built a fine, redundant network.
And others are willing to buy from you.
• Setting up neighbor sessions is the easy
part.
• Determine your policy.
• Filter EVERY route you hear from a
downstream customer, no matter how they
bitch. Can you say “blackholio”?

• You’ll need to add all customer prefixes to
that access-list 55 you use for controlling
your own outbound.
• You’ll need to keep modifying as-path
access-list 3 to add the ASNs of your
downstreams.
• This gets tedious.

ip as acc 3 permit ^$
ip as acc 3 permit ^22111$
ip as acc 3 permit ^22111 7007 30111$
etc...

• Doesn’t scale at some point.
• Solution? Communities...

Hot Potato
vs.
Hop-by-Hop
Routing

Hot Potato Routing
• Most providers try to hand off traffic
destined to the outside as soon as possible.
• Even if someone else’s network is bad.
• This will typically be the default behavior -
IF you re-set the MED on inbound routes.
• If not, you will do “Cold Potato” routing -
sending routes cross-country on your
network for UUNET’s other-coastal custs.

Quickie on Route-Maps
• Route-maps are used to match and set
attributes of routes. They are a little logic
flow of ANDs and NOT ANDs.
• Like a little basic program; evaluated in
order of the sequence number.
• At the end of evaluation, if a route has been
permitted at some point, it passed.
• A route-map is ADDITIVE to other filters.

Implementing Communities
ip comm 4 permit 4969:123


route-map tosprint deny 20
match comm 20
route-map tosprint permit 21
match comm 21
set as pre 4969
match comm 22
set as pre 4969 4969
match comm 4

Implementing Communities
route-map set-transit
match ip address 40
set comm 4969:1200 4969:666 additive

router bgp 22222
neigh <custip> route-map set-transit in

TUNING INBOUND
BGP ANNOUNCEMENTS

Inbound BGP Routes
• Inbound BGP routes make traffic go out.
Having a route means that an outbound
packet can use it as the basis for a
forwarding decision (well, the router can).
• It is far easier to adjust outbound routing
than inbound.
• Goal is generally to provide fastest, lowest-
loss, path for all destinations.

Tuning Inbound BGP Routes
• Policy
– Generally, to optimize throughput and latency.
– Could be to squash traffic to certain providers,
though, depending on the time of night and
state of mind of the network engineer in
question.
– Or, to reduce transit cost.
– Generally, though, it is to optimize connectivity
“quality”, whatever that is.

• Many destinations that you tune make
themselves known in the form of customer
complaints.
• Otherwise, start focusing on the biggest
providers (Sprint, UU, MCI/SW, ANS,
ATT, BBN, ...).

• Use traceroutes to determine connectivity.
• However, do the traceroute from the source
IP of the provider you are testing.
• No problem - do it from the border router
and the source IP will be that of the serial
interface.
• So, just set a temporary static route to a
given destination and trace away...

• Once you identify better paths, use
AS_PATH padding.
• Identify the providers in question.
• Pick out the relevant AS_PATH regexp.
• Build a route-map to apply inbound.

• Simple route-map
ip as acc 20 permit ^701 1673_
route-map inbound-uu permit 10
match as 20
set as pre 701 701
route-map inbound-uu permit 20
match as 1
• Always best to leave a specific match all at
the end.

• Other methods:

• We’ll talk about local_prefs later on...

TUNING OUTBOUND
BGP ANNOUNCEMENTS

Tuning Outbound BGP
• This is harder, because all of the other
networks implementing their own policies
complicate your life.
• Your two main tools are:
– Padding your outbound AS_PATHs
– Deaggregating announcements
• And:
– With a cooperative provider, using
communities

Tuning Outbound - Padding
• When your router announces iBGP routes,
it normally creates a 1-entry AS_PATH
with your ASN. So, by adding one or more
copy of your own ASN, you cause the
providers who listen to that route to de-
prefer it a bit (since the AS_PATH is now 1
longer, thus making it win less often).

Tuning Outbound - Padding
route-map pad-me-once
match as 1
set as prepend 22222

router bgp 22222
neigh 207.106.2.45 route-map pad-me-once out

Tuning Outbound - Communities
• If your providers are good (netaxs,
above.net, some others), they’ll give you
the ability to control your destiny with
communities.
• For example, netaxs honors the
communities:

Tuning Outbound - deagg.
• I have 207.106.128.0/17.
I want to advertise 207.106.128.0/17 to spr
and uu, and 207.106.128.0/18 to spr alone.
access 56 deny 207.106.128.0 0.0.63.255
access 56 <insert lines from access 55>
neigh <uunetip> dist 56 out

Peering
• Networks should peer as widely as possible,
for better interconnectivity. The more wide
the peering, the more traffic you will use.
This is a GOOD thing.
• Particularly, peering with local providers is
a very good thing.
• If you are both in a frame or SMDS cloud,
or in a room, peering makes sense...

BGP: Next-hop-self
• By default, our friend Mr. Promiscuous Q. eBGP
will pass on next-hops as imported
• So, if multiple routers are at a common XP, and one
party is transitting another, you might accidentally
send routes to a 3rd party which would cause them
to send traffic to your transit customer instead of
“bouncing” the packet off of you.
• Some people think transit over XPs is bad, but many
do it as a backup-of-last-resort. If you do, watch
your next-hops.

Next-Hop-Self Issues: Bad

192.41.177.241

192.41.177.87 AS 701
XP
AS 4969

AS 6666 says to AS 4969:
Prefix AS-Path Next-hop
10.10.10.0/24 6666 192.41.177.4
192.41.177.4
Prefix AS-Path Next-hop AS 6666
10.10.10.0/24 4969 6666 192.41.177.4

Next-Hop-Self Issues: Better

192.41.177.241

192.41.177.87 AS 701
XP
AS 4969

Prefix AS-Path Next-hop
10.10.10.0/24 6666 192.41.177.4
192.41.177.4
Prefix AS-Path Next-hop AS 6666
10.10.10.0/24 4969 6666 192.41.177.87

Stable BGP
• Nail routes to loopback.
• Watch out for flapping routes.
• Sites think that if a site shows instability, it
is worth blackholing for some time (30-90
minutes) until it stabilizes.
• Dampening hurts.
• How to escape from being dampened once
the underlying problem is fixed.

Stable BGP - Loopbacks
• Peering between loopbacks enhances
stability, since loopbacks don’t go down.
• Also, good for load-balancing.
• Set up lo0, then
• “neigh x.y.z.q update-source looback0”

BGP Stability - soft-reconfig
• Instead of hammering a session to cause
reevaluation (“clear ip bgp” drops the TCP
session), “clear ip bgp soft” can be used.
• “clear ip bgp x.y.z.q soft out” is low cpu; it
issues withdrawls for all currently-
advertised routes and recomputes and re-
sends roues.
• “clear ip bgp x.y.z.q soft in” is high
memory, as it needs to keep copy of all
routes received.

BGP Dampening
BGP dampening is used to minimize instability caused by route flapping and oscillation
over the network. To accomplish this, values are defined to identify badly behaved routes.
Genuity currently uses Cisco's default values which basically say 'flap 3 times in 15
minutes and your damped':

half-life 15 minutes
penalty 1000
suppress-value 2000
reuse-value 750

But in time it can become a bit more complex. Each time a route flaps it gets a penalty
(1000). If the cumulative penalty is greater than the suppress-value (2000, so 3 flaps), the
advertisement of the route will be suppressed. The penalty will be exponentially decayed
based on the half-life. Once the penalty decreases below the predefined reuse-value, the
advertisement will be unsuppressed. The default maximum suppress time is 4 times the
half-life. This ensures that if a route flaps a lot in a very short period but quickly becomes
stable it's only suppressed for a maximum of 1 hour.

Route flap dampening is not applied to routes originated from the AS in which the router
resides.

Why eBGP Multihop is BAD
• Why do it?
– “I can’t afford a router to take full routes in the middle”
– “Tough luck. Find another job or get bought so you can”
• What are we talking about?
– [7010, 64mb] <-> [2501, 16mb] <-> [4700, 64mb]
– 2501 defaults to 7010
– 7010 and 4700 speak eBGP multihop
• If you must:
– neigh <remoteip> ebgp 5 [# of hops]
– REALIZE that you’re breaking one of the Fundamental Postulates
of Active Routing - that every box in the middle knows (within a
few seconds) how to get to a destination because it HAD to know in
order to tell the box connected to it
• If someone puts a /24 route towards the 4700… Oh no..

Blackhole w/ eBGP Multihop
• One use - getting the real-time spam
blackhole feed from Vixie and company.
• Take the routes from a remote site and set
the next-hop on them to null0.
• Or, take the routes and set next-hop to one
of your upstreams to avoid the blackhole
effect.

SUPPORTING
MULTI-HOMED
CUSTOMERS

Supporting Multi-Homed Custs
• What they need from you is routes to the
‘net, and some ability to be flexible in how
they announce their routes.
• Routes to the ‘net - give them your
communities (“neighbor x.y.z.q send-
communities”). Publish your communities
so they know what they mean. WARN if
you change community semantics.

Supporting Multi-Homed Custs
• Be prepared to punch holes in your
aggregates.
– Using network statements, no problem.
– Otherwise, be prepared to use aggregate-
address.
• Set up communities they can use to control
which pipes you advertise them to.

BGP Confederations
• Or “BGP done right”
• Makes iBGP more promiscuous
• How?
– Fully-mesh all BGP speakers at a POP
– Use fake ASNs at each POP
– Between POPs, use eBGP rules (send everything)
– Within POPs, use iBGP rules
– Preserve local_prefs between POPs
• ‘bgp confed identifier 4969’
• ‘bgp confed peers 64512 64513 64514 64515’
• put in extra confed peers up-front

Logical View of full 16-router Mesh

(kudos to danny@genuity)

Confederations
C
C
C AS 64512
AS 1239 AS 701

C
C
C
C

C AS 64514
AS 64513
C C
C
C C

AS 4969

AS-Path filters for confederations

– ^$ Doesn’t work any more…
– ^$ matches internal routes, but with
confederations your routes will look like:
– ^(64512 64513)$ as well as ^$
– ip as acc 55 deny ^(([0-9 ]*))*$

Route Reflector Terminology

Client is used to identify “client” of the RR(s).
Non-client identifies standard BGP peers.
Cluster is a group of clients under same RR(s).
Cluster-id unique identifier for a cluster.
Originator-id router-id of the originator of the route.

Access-List 112: smd
– Deny 10.0.0.0/8, 172.16.0.0/16, 192.168.0.0/24
– Permit <= /24s in 192/8-205/8.
– Permit <= /19 in >= 206/8
– Permit /16 in the old B-space
– Permit <= /8 in the old A-space
– Deny everything else

Bigbgp (1)

More Related Content

What's hot (20)

Similar to Bigbgp (1) (20)

Bigbgp (1)