1. Policy-based Virtual Network Embedding across Multiple Domains
PolyViNE
Policy-based Virtual Network Embedding across Multiple Domains
Good afternoon everyone
Mosharaf Chowdhury
Fady Samuel, Raouf Boutaba
UC Berkeley
University of Waterloo

2. Virtual network embedding
is NP-hard 80 55 15 A B 10 22 a 12 c a a 90 50 10 12 10 20 C D E F b b c c 15 10 60 85 10 10 27 25 b
VN Request (SP) G H 17 70 65
Substrate Network (InP)

3. Inter-domain ViNE D A C B

4. Bird’s-eye-view solution
Partitioning the VN request into K components
Embedding individual components into K substrate networks
Establishing inter-connection between them

5. Major challenges How do InPs and SPs find each other?
Framework for resource trading for rapid VN instantiation and fair value
Without sacrificing local autonomy
Tussles between contrasting utility functions
InPs trying to maximize individual profits
SPs trying to get the best price
Secrecy and privacy concerns of the InPs
Information on internal topology, resources etc.
How do InPs and SPs find each other?
How to ensure the best price?
How to share information?

6. Approaches Full disclosure Third-party Minimal disclosure
Publicly available InP information
Third-party
InP information must be shared with the broker
Possibility of monopoly (trust issues)
Minimal disclosure
No central entity
Safest of the three, but the hardest as well

7. PolyViNE design choices
Decentralized embedding
No central entity with knowledge of internal policies
Local autonomy with global competition
InPs are free to choose individual policies and embedding algorithms
Competitive pricing at every stage of embedding
Location assisted embedding
Guided by the location constraints on virtual nodes and the location information of the substrate nodes

8. Workflow summary
InP #1
InP #6
InP #4
InP #7 SP
InP #2
InP #5
InP #3

9. Whether & how to embed locally?
InP workflow SP
InP #1
InP #2
InP #3
InP #4
InP #5
InP #6
InP #7
Whether & how to embed locally?
How to forward?
Where to forward?

10. Location assisted forwarding
Informed request forwarding
Minimize flooding
Avoid unnecessary random forwarding
Two components
Hierarchical addressing scheme (COST)
Location awareness protocol (LAP)

11. COST Hierarchical addressing scheme Continent.cOuntry.State.ciTy
Allows prefix aggregation
Provides high flexibility in expressing virtual node location constraints
Allows InPs to obfuscate topology information
Continent.cOuntry.State.ciTy
NA.CA.ON.Toronto: Node in Toronto
NA.CA.ON.*: Node anywhere in Ontario

12. LAP: Location Awareness Protocol
InPs exchange LAP updates to build local policy compliant view of the InP network
Each entry of an InP’s LAP database contains a mapping from a COST prefix to a set of paths to InPs with that prefix
Each path has an associated estimated price

13. LAP Resource prices can rapidly fluctuate in a dynamic environment
Gossip is too slow to propagate price updates
Staleness
Use a hybrid of Gossip and Publish/Subscribe
InPs can get direct and frequent updates

14. Simulation ViNE algorithms are hard to evaluate Simulation settings
What would be a representative input dataset?
Which are the best metrics & how to measure them?
Only look into simple convergence characteristics
Simulation settings
Based on settings used in existing intra-domain work
100 InPs with random links between them, each with nodes and links
Max recursive probe depth set to 12

15. How many InPs must collaborate?
60%

16. Summary PolyViNE is a policy-based inter-domain VN embedding framework
Local autonomy with global competition
Decentralized location-assisted embedding using COST and LAP
Possible future work (among many)
Interaction between diverse local ViNE algorithm
Game-theoretic analysis of the proposed scheme

17. Thanks! ? ? ?

18. Backup