1 / 17

Resolve the Virtual Network Embedding Problem: A Column Generation Approach

This paper discusses network virtualization, the virtual network embedding problem, and proposes a column generation approach for optimal solution. It also compares the approach with other methods and evaluates its performance.

johnstond
Download Presentation

Resolve the Virtual Network Embedding Problem: A Column Generation Approach

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Resolve the Virtual Network Embedding Problem: A Column Generation Approach Qian Hu, Yang Wang, Xiaojun Cao Department of Computer Science Georgia State University Atlanta, GA, 30303

  2. What is Network Virtualization? • ISP Decoupling: Infrastructure Providers (InPs) and Service Providers (SPs) • InPs: manage the physical infrastructure • SPs: operate virtual network, offer E2E user services • A programmable infrastructure • InPs resources efficient utilization • Hardware, energy, recovery • SPs can deploy services fast • No high initial investments on the infrastructures • Flexible deployment • Disruptive technologies deployed by InPs will not affect supported services

  3. Virtual Network Embedding Problem Given the virtual network (VN) request GV = (NV , LV ), and substrate network GS = (NS , LS ), Objective: Map the VN with least cost Constraints: Node Computing: One virtual node mapped onto one substrate node No two virtual nodes share the same substrate node; Link/path bandwidth capacity One virtual link mapped to substrate links or path(s) Others such as location, protection

  4. Bridging SP and InP: Virtual Network Embedding Virtual Network: the logical network of SP 4 8 5 8 6 3 Substrate Network: the physical network of InP 20 2 20 20 5 26 30 20 40 30 20

  5. Literature Work Virtual Network Embedding (VNE): NP-Complete Optimal solutions from link-based ILP [Chowdhury etc, INFOCOM’09] Extensive computational time Not practical Heuristic approaches Relaxation of link-based ILP [Chowdhury etc, INFOCOM ’09] Others[e.g., Lischka etc, ACM VISA’09] Not optimal Not sure how far away from optimal

  6. Major Contributions of Our Work Propose path-based ILP model for VNE problem Propose a column generation process Integrated with a branch-and-bound framework Resolve the VNE problem optimally in practice Obtain sub-optimal results with guaranteed performance (with branch-and-bound)

  7. Network Embedding: Network Flow Model Virtual Network 1 1 Virtual Network Embedding => Multi-commodity Flow problem a a 2 2 3 3 c b c b c 5 4 4 5 1 Substrate Network 2 3 a 4 5 auxiliary edge: connect a virtual node to eligible substrate nodes b

  8. Path-based ILP Formulation Amount of Flow on Path p Exponential number of paths c Link is not overloaded 1 2 3 Xa,2 Each commodity is satisfied Xa,4 5 a Node Assignment 4 Traffic only on the link to the “mapped” node Primal b

  9. Path-based Formulation-Primal and Dual :ye :λk :yi :yI exponential # path choices => exponential # of constraints :ΠI,i Primal Dual Fact 1: OPT(P) = OPT(D-P)

  10. Column Generation to try path selection PO: set of optimal paths P’: path space we look at P: potential exponential path space P’ Target: OPT(P’) =OPT(P) PO P P’ PO Q1: Is P’ optimal? Q2: Which path to add to P’ ? P

  11. Q1: P’ optimal? & Q2: which path to add to P’ Exponential # of Constraints (D-P) Solution with some Constraints (D-P’) OPT(D-P’) P’ Q3: Check Exp. # of constraints? Feasible for other constraints in P? Q1 PO No Add the constraint’s corresponding path to P’: Q2 P Yes OPT(D-P’) =OPT(D-P) OPT(P’) = OPT(P) OPT(D-P’) = OPT(P’) >= OPT(P) = OPT(D-P)

  12. Check all the Constraints: Shortest Path Problem Weight for auxiliary edge WI,i=ΠI,i c Wc-1 Weight for substrate links We = ye+ce Wc-3 1 W1-2 W1-3 Wa-2 Wa-2+W1-2+Wc-1≥λk W1-4 2 3 a W3-4 W2-4 W3-5 Wa-4 W4-5 4 5 Rational: if the shortest path of each commodity satisfies, all the paths satisfy. Wb-5 Wb-4 b

  13. Overall Framework Theorem 1: The process to identify a set of optimal paths is Polynomial. Obtain a subset P’ Solve the D-P’ Solution feasible for D-P NO Increase P’ YES Solution for D-P Found

  14. Performance Evaluation Simulation Setting Virtual network node number [2-10] Substrate network node number [10-50] Average connectivity 50% Virtual network link/node capacity [1-20] Substrate network link/node capacity [1-50] Compared Approaches Link-based ILP (from prior work Infocom’09) Path-based ILP ILP P-VNE Relaxed Path-based ILP ILP P-VNE’ k=1 ILP P-VNE’ k=2 ILP P-VNE’ k=3

  15. Optimality Comparison Both P-VNE and Link-based ILP achieve optimality Increasing k improves the performance for relaxed P-VNE P-VNE has considerable less computational time than Link-based ILP Increasing k also increases time Over small k may leads to infeasible solution Resource Consumption (ce=1) QoS (ce=latency of e)

  16. Summary What is Network Virtualization? Virtual Network Embedding and Network Flow Model Path-based VNE Model Column Generation Approach

  17. Questions? Resolve the Virtual Network Embedding Problem: A Column Generation Approach Qian Hu, Yang Wang, Xiaojun Cao Thank you!

More Related