Wen Xu and Jennifer Rexford

1
MIRO Multi-path Interdomain ROuting

• Wen Xu and Jennifer Rexford
• Princeton University

2
If Some if Good, More is Better

• More flexible path selection
• Avoid a path with an undesirable hop
• Better load balancing
• Split traffic over multiple paths
• Faster recovery from failure
• Switch from one path to another
• Secure routing
• Avoid paths controlled byan adversary

B
C
A
F
D
E
3
So, Why Not do Source Routing?

• The ultimate in flexibility
• Sender determines path for each packet
• At the cost of
• Lost control for intermediate ASes
• Propagating topology information

B
C
A
ABEF
F
D
E
ADECF
4
Hmm, Maybe BGP Isnt So Bad

• Internet is a big, federated network
• Local policies for path selection export
• Efficient propagation of path information
• Add multi-path to BGP
• AS-level, path-vector routing
• Extended for multi-path

5
Pull-Based Negotiation
BEFBCF
CFCEFCBEF
B
C
ABEFADEF
A
F
F
D
E
EF ECF
DEFDABEF

• Pull-based route retrieval
• Solicit routes only when necessary
• Bilateral negotiations
• AS relationships usually bilateral anyway

6
Not Just Your Neighbors
BEFBCF
CFCEFCBEF
ABEFADEF
B
C
F
A
F
D
E
DEFDABEF
EF ECF

• The two ASes might not be neighbors
• Either AS can initiate the negotiation

7
IP-in-IP Encapsulation
e
e
d
d
B
C
A
F
D
E

• Direct packet along alternate route
• Destination-based forwarding not enough
• Encapsulate the packet to egress point

8
Flexibility, Control, and Scalability

• Flexibility for edge ASes
• AS can learn and use multiple paths
• Control for intermediate ASes
• Selective export of extra routes
• Scalability of the global system
• AS-level routing protocol
• Pull-based route retrieval
• Selective export of routes

9
Evaluation Methodology

• Infer AS graph and relationships
• RouteViews BGP data
• Gao inference algorithm
• Customer, peer, and provider relationships
• Three variations on export rules
• Strict policy all paths in same class
• Respect export all policy-compliant paths
• All export all of the paths

10
Evaluation Methodology (Cont.)

• Who to negotiate with?
• Neighboring ASes
• ASes along the default path
• How many ASes initiate negotiation?
• Only sending AS
• Only one IP-in-IP encapsulation
• What path-selection policies?
• Avoid a particular AS
• Inbound traffic engineering

11
Avoiding an AS Success Rate
Date BGP MIRO/s MIRO/e MIRO/a SourceRouting
2000 27.8 65.4 72.9 75.3 89.5
2003 31.2 67.0 74.6 76.6 90.4
2005 29.5 67.8 73.7 76.0 91.1
Legacy BGP at 30
MIRO at 70-75
Source routing only 10-15 better
12
Avoiding AS Overhead (for 2005)
Policy Success rate AS /tuple Path /tuple
Strict 67.8 2.80 36.6
Export 73.7 2.53 58.9
Flexible 76.0 2.38 139.0

• With more flexible policies
• Negotiate with fewer ASes
• But learn more paths from each AS

13
Avoiding AS Incremental Deployment
99.9 of total gain if 25 of nodes adopted MIRO
53 of total gain if 0.2 of nodes (40 nodes)
adopted MIRO
82 of total gain if 25 of nodes adopted MIRO
44 of total gain if 0.2 of nodes (40 nodes)
adopted MIRO
14
Conclusions

• Multi-path extension to BGP
• Retain AS-level, path-vector routing
• Add pull-based route retrieval
• and IP-in-IP encapsulation
• Evaluation results
• Satisfies the ASs policy goals
• Avoids state explosion
• Offers benefits in small deployments

15
Making MIRO a Reality

• Data plane
• Encapsulation and directed forwarding
• Preventing unauthorized packet deflection
• Control plane
• Monitoring the BGP-learned routes
• Disseminating the alternate paths
• Financial plane
• ISPs offering multi-path as a paid service