1 Interdomain Traffic Engineering with BGP By Behzad Akbari Spring 2011 These slides are based on the slides of Tim. G. Griffin (AT&T) and Shivkumar (RPI)

2 Real World: Multiple Links Between Domains Client Web server Multiple links Middle of path

3 BGP Decision Process Highest local preference Lowest AS path length Lowest MED (with same next hop AS) I-BGP < E-BGP Lowest IGP cost to next hop Lowest router ID of BGP speaker

4 BGP Route Selection Process Prefer largest LOCAL-PREF If same LOCAL-PREF prefer the shortest AS-PATH. If AS-PATHs are the same prefer the lowest MED. If MED is same, prefer min-cost NEXT-HOP If routes learned from EBGP or IBGP, prefer paths learnt from EBGP Final tie-break: Prefer the route with I-BGP ID (IP address)

5 Route Selection Summary Highest Local Preference Shortest ASPATH Lowest MED i-BGP < e-BGP Lowest IGP cost to BGP egress Lowest router ID traffic engineering Enforce relationships

6 Hot-Potato Routing San Francisco Dallas New York Hot-potato routing = route to closest egress point when there is more than one route to destination ISP network 9 10 dest multiple egress points -All traffic from customer to peers -All traffic to customer prefixes with multiple connections

7 Hot-Potato Routing Change San Francisco Dallas New York ISP network dest failure - planned maintenance - traffic engineering 11 Routes to thousands of destinations switch egress points!!! Consequences:  Transient forwarding instability  Traffic shift  Interdomain routing changes 11

8 Tuning BGP to control the outgoing traffic Principle  To control its outgoing traffic, a domain must tune the BGP decision process on its own routers How to tune the BGP decision process ? Filter some routes learned from some peers  local-pref usual method of enforcing economical relationships  MED usually, MED value is set when sending a route but some routers allow to insert a MED in a received route  allows to prefer routes over others with same AS Path length  IGP cost to nexthop setting of IGP cost for intradomain traffic engineering several routes in forwarding table instead of one

9 Local Preference AS1 AS2 MED Load-Balancing Knobs in BGP LOCAL-PREF: outbound traffic, local preference (box-level knob) MED: Inbound-traffic, typically from the same ISP (link-level knob)

10 Local Preference Attribute Local to AS  It is never advertised to an eBGP peer. Used to influence BGP path selection AS /24 AB / Preferred by all AS3847 routers /24 80 AS 3847 G FE CD Default 100  Highest local-pref preferred For example, you can express the policy “prefer private connect” by making the “local_pref” be 150 and leaving all other peers at 100.

11 Controlling incoming traffic by outbound BGP routes Outbound BGP routes make traffic come in  It’s a lot harder to control inbound traffic as other ASs’ policies complicate your life! If you are a stub AS with a single connection  Not much you need to do except to filter out routes not in your AS If you are a multi-homed stub AS,  Want to control through which link/provider that traffic to certain destinations in your AS may take, to load balance or for back-up If you are an ISP, you want to minimize transit cost,  carry transit traffic from customers only !  use “hot-potato” routing to hand off traffic to peers/providers as soon as possible  to load balance, or to ensure reliability with back-up routes

12 Why Inbound Traffic is Hard to Manage Other ASes decide how to send to you  Destination-based routing  Other ASes decide which path to take  Based on their own policies AS 2 doesn’t know how AS 1 will send traffic toward p p

13 Tuning BGP to control the incoming traffic Principle  To control its incoming traffic, a domain must tune the BGP advertisements sent by its own routers How to tune the BGP advertisements ?  Do not announce some routes to from some peers advertise some prefixes only to some peers  MED insert MED=IGP cost, usually requires bilateral agreement  AS-Path artificially increase the length of AS-Path

14 AS Prepending Artificial increasing AS path length  Prepend your own AS in the path  E.g., turn “3 4 5” into “ ”  Hope to make the path less attractive 1 3 “ ” “3 4 5”

15 ASPATH Padding: Shed inbound traffic Padding will (usually) force inbound traffic from AS 1 to take primary link AS /24 ASPATH = customer AS 2 provider /24 backupprimary /24 ASPATH = 2

16 Padding May Not Shut Off All Traffic AS /24 ASPATH = customer AS 2 provider /24 ASPATH = 2 AS 3 provider AS 3 will send traffic on “backup” link because it prefers customer routes and local preference is considered before ASPATH length! Padding in this way is often used as a form of load balancing backupprimary

17 Multiple Exit Discriminator (MED) Tell your neighbor what you want  MED attribute to indicate receiver preference  Decision process picks route with smallest MED  Can use MED for “cold potato” routing  But, have to get your neighbor to accept MEDs 1 3 “3 4 5” with MED=2 “3 4 5” with MED=1

18 Hot Potato Routing: Closest Egress Point / IGP distances egress 1 egress 2 This Router has two BGP routes to /24. Hot potato: get traffic off of your network as Soon as possible. Go for egress 1!

19 Getting Burned by the Hot Potato High bandwidth Provider backbone Low bandwidth customer backbone Heavy Content Web Farm tiny http request huge http reply SFFNYC San Diego

20 Cold Potato Routing with MEDs (Multi-Exit Discriminator Attribute) Heavy Content Web Farm /24 MED = /24 MED = 56 This means that MEDs must be considered BEFORE IGP distance! Prefer lower MED values Note1 : some providers will not listen to MEDs Note2 : MEDs need not be tied to IGP distance

21 MEDs Can Export Internal Instability Heavy Content Web Farm /24 MED = /24 MED = 56 OR FLAP