Selfish Behavior in Interdomain Routing

1
Selfish Behavior in Inter-domain Routing
"Implications of Autonomy for the Expressiveness
of Policy Routing", Nick Feamster, Ramesh Johari,
Hari Balakrishnan. SIGCOMM 2005. "Routing and
peering in a competitive Internet, Johari, R.,
and Tsitsiklis, J.N. (2003). LIDS Publication
2570.

• Presented by Weifeng Chen

2
Implications of Autonomy for the Expressiveness
of Policy Routing

• Nick Feamster, MIT
• Ramesh Johari, Stanford
• Hari Balakrishnan, MIT

3
Internet Routing is Policy-Based
Abilene
Comcast
MIT
ATT
Cogent
Expressiveness ASes can achieve
objectivesAutonomy ASes can configure policies
independently
Expressiveness vs. autonomy
4
Talk Outline

• Background
• Expressiveness ranking and filtering
• Autonomy independent ASes
• Stability and Safety
• Filtering advertise paths to other AS
• Ranking and safety
• Dispute rings and safety
• Autonomy and safety

5
Expressiveness Ranking and Filtering
Ranking route selection
Customer
Primary
Competitor
Backup
Ranking controls traffic out of the network
traffic engineering Filtering controls
traffic into the network business contracts
Must have autonomy!
6
Preferred and consistent paths

• Routing system
• Paths from ASes to common destination 0
• Preference a total order
• Prefer P to Q
• Feasible specified by filtering
• Consistent paths aligned with feasibility
• E.g, if ijP0 is a path from i to 0, then P must
  be a feasible path from j to 0

7
Stable and safe routing system

• Discrete activation of AS
• Single AS at each time step
• Process all incoming advertisement and make
  decision
• Stable
• All paths are preferred
• Safe
• The routing paths do not oscillating when ASes
  activate, given any initial path assignment

Under ranking and filtering, with autonomy
8
Goal Local Constraints for Safety
Given no restrictions on filtering or topology,
what are the local restrictions on rankings to
guarantee global safety under filtering?
9
Safety No Persistent Oscillation
Depends on the interactions of rankings and
filters of multiple ASes
1 3 0 1 0
0
2 1 0 2 0
3 2 0 3 0
Dispute wheel global, cyclic relationship among
rankings
Varadhan, Govindan, Estrin, Persistent Route
Oscillations in Interdomain Routing, 1996
Griffin, Shepherd, Wilfong, The Stable Paths
Problem and Interdomain Routing, ToN, 2002
10
Contribution 1

• Next-hop ranking
• Preference only depends on next hop
• Known result If no filtering, using next-hop
  ranking, there exists a stable path assignments
• New results
• A routing system where each node has only a
  next-hop ranking may not be safe
• Introducing filtering can yield a system where no
  stable path assignment exists

11
Example
3,0,2
Filter 3 from 2
0
Filter 1 from 3
Filter 2 from 1
1,0,3
2,0,1
12
Contribution 2 First Necessary Condition for
Safety

• Known result No dispute wheel ? Safe

Safe
No Dispute Ring

• New result
• No dispute wheel? Safe under filtering
• Dispute ring implies no safety under filtering

Safe under Filtering
No Dispute Wheel
Dispute ring Dispute wheel where each node only
appears once
Problem No dispute ring is still a global
condition.
13
Contribution 3 Autonomous

• Providing complete autonomy and filtering,
  shortest paths ranking guarantee safety

ARC Function
Rankings (from single AS)
Accept/Reject
14
ARC Function Properties
Permutation Invariance Node labels dont matter
ARC Function
Accept
Accept
Scale Invariance
Adding new nodes does not force a node to change
its rankings over old paths.
15
Examples of ARC Functions

• Accept only next-hop rankings
• Captures most routing policies
• Problem system may not be safe
• (See Section 4.1 for proof)
• Accept only shortest hop count rankings
• Guarantees safety under filtering
• Problem not expressive

16
What ARC Functions Violate Safety?
Theorem. Permitting paths of length n2 over
paths of length n will violate safety under
filtering. Theorem. Permitting paths of length
n1 over paths of length n will result in a
dispute wheel.
17
Conclusion

• Factoring expressiveness ranking and filtering
• Dispute Ring First necessary condition for
  safety
• ARC Function Abstraction for autonomy
• Restrictions on rankings that guarantee safety
  and preserve autonomy

18
Routing and Peering in a Competitive Internet

• Ramesh Johari, MIT
• John Tsitsiklis, MIT

19
Growth of Internet

• Before 1995
• Maintained and directed by government and
  academia
• Overall performance best interested of all
  parties
• But after 1995
• Loosely federation of independent network
  providers
• Individual objectives not necessarily aligned
• Self interested behavior to maximize own profits

Goal Understand economic incentives driving
actions of network providers
20
Nearest exit in peer networks

• Peer networks B will only accept traffic from A
  that is destined for points within B, and vice
  versa.
• Cost only network routing costs
• Nearest exit or hot potato outgoing traffic
  exits a provider's network as quickly as possible

21
Optimal replacement of interconnection

• Under nearest exit routing, where two peer
  networks choose to interconnect?

• NP-complete in general
• Sender find a set of interconnecting points that
  minimize the cost of sending traffic to the
  receiver network
• Special cases (sender and receiver have a linear
  or tree topology), there exists a unique peering
  point placement which will simultaneously satisfy
  both

22
Nearest exit vs. Optimal Routing

• Providing interconnection replacement, nearest
  exit routing is not optimal, according to a
  priori network cost metric
• But if network cost is measured by assessing a
  cost per unit flow traversing each link, then
  cost using nearest exit is no worse than three
  times the optimal (shortest path)