1. Part II: Inter-domain Routing Policies

2. March 8, 20042 What is routing policy? ISP1 ISP4ISP3 Cust1Cust2 ISP2 traffic Connectivity DOES NOT imply reachability! Policy determines how traffic can flow on the Internet

3. March 8, 20043 BGP routing process Apply input policy Routes received from peers Select best route Best routes Apply output policy Routes advised to peers Routing table Forwarding table BGP is not shortest path routing!

4. March 8, 20044 Best route selection  Highest local preference  Shortest AS path  Lowest MED  I-BGP < E-BGP  Lowest I-BGP cost to E-BGP egress  Tie breaking rules

5. March 8, 20045 Best route selection  Highest local preference  To enforce economical relationships between domains  Shortest AS path  Lowest MED  I-BGP < E-BGP  Lowest I-BGP cost to E-BGP egress  Tie breaking rules

6. March 8, 20046 Best route selection  Highest local preference  Shortest AS path  Compare the quality of routes, assuming shorter AS-path length is better  Lowest MED  I-BGP < E-BGP  Lowest I-BGP cost to E-BGP egress  Tie breaking rules

7. March 8, 20047 Best route selection  Highest local preference  Shortest AS path  Lowest MED  To implement “cold potato” routing between neighboring domains  I-BGP < E-BGP  Lowest I-BGP cost to E-BGP egress  Tie breaking rules

8. March 8, 20048 Best route selection  Highest local preference  Shortest AS path  Lowest MED  I-BGP < E-BGP  Prefer EBGP routes to IBGP routes  Lowest I-BGP cost to E-BGP egress  Tie breaking rules

9. March 8, 20049 Best route selection  Highest local preference  Shortest AS path  Lowest MED  I-BGP < E-BGP  Lowest I-BGP cost to E-BGP egress  Prefer routes via the nearest IGP neighbor  To implement “hot potato” routing  Tie breaking rules

10. March 8, 200410 Best route selection  Highest local preference  Shortest AS path  Lowest MED  I-BGP < E-BGP  Lowest I-BGP cost to E-BGP egress  Tie breaking rules  Router ID based: lowest router ID  Age based: oldest route

11. March 8, 200411 BGP route propagation  Not all possible routes propagate  Commercial relationships determine policies for  Route import  Route selection  Route export

12. March 8, 200412 Typical AS relationships  Provider-customer  customer pay money for transit  Peer-peer  typically exchange respective customers’ traffic for free  Siblings

13. March 8, 200413 Transit vs. peering  ISP definition  Internet service provider is an organization that sells access to the Internet  Transit definition  “Business relationship whereby one ISP provides (usually sells) access to all destinations in its routing table”.  Peering is non-transitive relationship  A peers with B, B peers with C, does not imply A peers with C

14. March 8, 200414 What is peering?  Peering definition  “An interconnection business relationship whereby ISPs provide connectivity to each others’ transit customers.”  Hybrid exists  Regional transit  Paid peering

15. March 8, 200415 Example of commercial relationship UMich Merit Cogent GoogleBerkeley ESnet

16. March 8, 200416 Tier-1 peering  Buy no transit from any other providers  Have only customers and peers  Has full mesh peering with other tier-1’s  Motivation for peering:  Minimize their interconnection costs while providing sufficient interconnection BW to support customer and its growth

17. March 8, 200417 Tier-2 peering  ISP that purchases (resells) transit within an Internet region  Benefits  Decreases the cost and reliance on purchased Internet transit  Lowers inter-AS traffic latency  Fewer AS hops, AS peering links traversed

18. March 8, 200418 Is peering always better than transit?  Concerns of peering  Traffic asymmetry  No SLAs: less liability or incentive to improve performance  “Free” rather than getting paid  Peers become more powerful

19. March 8, 200419 Where to peer?  Public peering: at public peering locations  Private peering  Exchange-based interconnection model  A meet point at which ISPs exchange traffic  Can be neutral Internet business exchange  Direct circuit interconnection model  Point-to-point circuit between the exchange parties

20. March 8, 200420 What are siblings?  Mutual transit agreement  Provide connectivity to the rest of the Internet for each other  Typically between two administrative domains such as small ISPs or universities located close to each other, cannot afford additional Internet services for better connectivity

21. March 8, 200421 AS relationships translate into BGP export rules  Export to a provider or a peer  Allowed: its routes and routes of its customers and siblings  Disallowed: routes learned from other providers or peers  Export to a customer or a sibling  Allowed: its routes, the routes of its customers and siblings, and routes learned from its providers and peers

22. March 8, 200422 Which AS paths are legal?  Valley-free:  After traversing a provider-customer or peer-peer edge, cannot traverse a customer-provider or peer-peer edge  Invalid path: >= 2 peer links, downhill- uphill, downhill-peer, peer-uphill

23. March 8, 200423 Example of valley-free paths X X [1 2 3], [1 2 6 3] are valley-free [1 4 3], [1 4 5 3] are not valley free

24. March 8, 200424 Inferring AS relationships  Identify the AS-level hierarchy of Internet  Not shortest path routing  Predict AS-level paths  Traffic engineering  Understand the Internet better  Correlate with and interpret BGP update  Identify BGP misconfigurations  E.g., errors in BGP export rules

25. March 8, 200425 Existing approaches  On inferring Autonomous Systems Relationships in the Internet, by L. Gao, IEEE Global Internet, 2000.  Characterizing the Internet hierarchy from multiple vantage points, by L. Subramanian, S. Agarwal, J. Rexford, and R. Katz, IEEE Infocom, 2002.  Computing the Types of the Relationships between Autonomous Systems, by G. Battista, M. Patrignani, and M. Pizzonia, IEEE Infocom, 2003.

26. March 8, 200426 Gao’s approach  Assumptions  Provider is typically larger than its customers  Two peers are typically of comparable size

27. March 8, 200427 Gao’s algorithm  Find the highest degree AS node to be the top provider of the AS path  Left to the top node  customer-provider or sibling-sibling links  Right to the top node  provider-customer or sibling-sibling links  Sibling-sibling  if providing mutual transit service for each other  Peer-peer  with top provider and of comparable degree value

28. March 8, 200428 Subramanian’s Approach  Use BGP tables from multiple vantage points  More complete  Exploit uniqueness of each point  Build AS-level hierarchy of Internet  Relationship based, not degree based  5 level classification of AS’s

29. March 8, 200429 Relationship inference rules  Position of AS in AS graph gives rank  Combine ranks from multiple tables  Compare ranks  Peer-peer with similar ranks  Provider-customer: provider with higher ranks

30. March 8, 200430 Hierarchy inference  Internet hierarchy inference  Based on relationships  Not degree [Gao]

31. March 8, 200431 Battista’s approach  Cast it as an optimization problem to find provider-customer relationships that minimize the number of conflicts  Shows the problem is NP-hard  Do not deal with peer-peer relationships well

32. March 8, 200432 Policy routing causes path inflation  End-to-end paths are significantly longer than necessary  Why?  Topology and routing policy choices within an ISP, between pairs of ISPs, and across the global Internet  Peering policies and interdomain routing lead to significant inflation  Interdomain path inflation is due to lack of BGP policy to provide convenient engineering of good paths across ISPs

33. March 8, 200433 Path inflation  Based on [Mahajan03]  Comparing actual Internet paths with hypothetical “direct” link