Upgrade to Pro — share decks privately, control downloads, hide ads and more …

MULE: Multi-Layer Virtual Network Embedding

MULE: Multi-Layer Virtual Network Embedding

Network Virtualization (NV), considered as a key enabler for overcoming the ossification of the Internet allows multiple heterogeneous virtual networks to co-exist over the same substrate network. Resource allocation problems in NV have been extensively studied for single layer substrates such as IP or Optical networks. However, little effort has been put to address the same problem for multi-layer IP-over-Optical networks. The increasing popularity of multi-layer networks for deploying backbones combined with their unique characteristics (e.g., topological flexibility of the IP layer) calls for the need to carefully investigate the resource provisioning problems arising from their virtualization. In this paper, we address the problem of MUlti-Layer virtual network Embedding (MULE) on IP-overOptical networks. We propose two solutions to MULE: an Integer Linear Program (ILP) formulation for the optimal solution and a heuristic to address the computational complexity of the optimal solution. We demonstrate through extensive simulations that on average our heuristic performs within ≈1.47× of optimal solution and incurs ≈66% less cost than the state-of-the-art heuristic.

Shihabur Rahman Chowdhury

November 30, 2017
Tweet

More Decks by Shihabur Rahman Chowdhury

Other Decks in Research

Transcript

  1. MULE: Multi-Layer Virtual Network Embedding Shihabur R. Chowdhury, Sara Ayoubi,

    Reaz Ahmed, Nashid Shahriar, Raouf Boutaba Jeebak Mitra, Liu Liu
  2. Virtual Network Embedding (VNE) 2 10 a b c 10

    10 10 12 10 d e f 20 20 20 5 5 C A B D E F G H 60 80 55 50 70 65 85 90 22 15 12 10 15 17 17 20 25 a b c e d f
  3. Virtual Network Embedding (VNE) 3 10 a b c 10

    10 10 12 10 d e f 20 20 20 5 5 C A B D E F G H 60 80 55 50 70 65 85 90 22 15 12 10 15 17 17 20 25 a b c e d f Extensive Literature, mostly focused on single-layer substrate
  4. Multi-Layer IP-over-Optical Network 4 A D E C B IP

    Network  Packet Switched  Flexible addressing, traffic engineering, resource allocation
  5. Multi-Layer IP-over-Optical Network 5 1 3 2 4 6 5

    7 9 8  Circuit switched  High capacity (Terabits of bandwidth/link) Optical Network
  6. Multi-Layer IP-over-Optical Network 6 A D E C B 1

    3 2 4 6 5 7 9 8 IP overlay on Optical Network  IP routers are directly connected to optical switches  IP links are logical and tunneled over optical paths  Best of two worlds  High capacity combined with flexible addressing, routing, traffic engineering, resource allocation.
  7. Multi-Layer IP-over-DWDM Network 8 A D E C B 1

    3 2 4 6 5 7 9 8 1 3 5 A D IP Links are tunneled over a single wavelength light-path
  8. Multi-Layer IP-over-OTN Network 9 A D E C B 1

    3 2 4 6 5 7 9 8 1 3 5 A D OTN Links are logical, routed over wavelengths, and can multiplex bandwidth of multiple IP Links
  9. Topological Flexibility of Multi-Layer Network 11 A D E C

    B 1 3 2 4 6 5 7 9 8 1 3 5 A D New IP Links can be created on-the-fly
  10. (One Possible) Answer: If IP network does not have sufficient

    capacity for VN embedding, then we can increase capacity, by creating new IP links 13
  11. The Problem Multi-Layer Virtual Network Embedding (MULE) 14 In the

    most resource efficient way, jointly determine
  12. The Problem Multi-Layer Virtual Network Embedding (MULE) 15 Creation of

    New IP links (if necessary) In the most resource efficient way, jointly determine A D E C B ?
  13. The Problem Multi-Layer Virtual Network Embedding (MULE) 16 Creation of

    New IP links (if necessary) VN Embedding on the IP Layer In the most resource efficient way, jointly determine d f e A D E C B ? A D E C B d f e
  14. The Problem Multi-Layer Virtual Network Embedding (MULE) 17 Creation of

    New IP links (if necessary) VN Embedding on the IP Layer Embedding of new IP Links on Optical Layer In the most resource efficient way, jointly determine d f e 1 3 2 4 6 5 7 9 8 A D E C B ? A D E C B d f e A D E C B
  15. Context 18 Multi-Layer IP-over-OTN Network OTN is static and OTN

    Links are already provisioned on light-paths in DWDM layer. No multi-path embedding; No node capacities
  16. MULE: Example 19 A D E C B 1 3

    2 4 6 5 7 9 8 15 10 10 10 15 1000 985 990 990 990 985 1000 Multi-Layer Substrate Network Given
  17. MULE: Example 20 A D E C B 1 3

    2 4 6 5 7 9 8 15 10 10 10 15 1000 985 990 990 990 985 1000 Multi-Layer Substrate Network Logical IP Layer Physical Optical Layer Given
  18. MULE: Example 21 A D E C B 1 3

    2 4 6 5 7 9 8 15 10 10 10 15 1000 985 990 990 990 985 1000 Multi-Layer Substrate Network Logical IP Layer Physical Optical Layer Given 0 2 1 {C} {A, B} {D, E} 15 15 15 Virtual Network (VN) Location Constraint
  19. MULE: Example 22 A D E C B 1 3

    2 4 6 5 7 9 8 15 10 10 10 15 1000 985 990 990 990 985 1000 Embed the VN on the IP Layer 0 2 1
  20. MULE: Example 23 A D E C B 1 3

    2 4 6 5 7 9 8 15 10 10 10 15 1000 985 990 990 990 985 1000 Create new IP links (if necessary) Embed the VN on the IP Layer 0 2 1
  21. MULE: Example 24 A D E C B 1 3

    2 4 6 5 7 9 8 15 10 10 10 15 1000 985 990 990 990 985 1000 Create new IP links (if necessary) Embed the VN on the IP Layer Embed the new IP links on Optical Layer 0 2 1
  22. MULE: Example 25 A D E C B 1 3

    2 4 6 5 7 9 8 15 10 10 10 15 1000 985 990 990 990 985 1000 Create new IP links (if necessary) Embed the VN on the IP Layer Embed the new IP links on Optical Layer 0 2 1 Objective: Minimize bandwidth allocation cost on both layers
  23. Our Contributions 26 OPT-MULE FAST-MULE ILP-based Optimal Solution (NP-hard) Three

    Step Heuristic: Collapse, Extract, Embed A suit of solutions to MULE
  24. State-of-the-art 27 No Optimal Solution ILP-based Optimal Solution Two step

    virtual node and virtual link embedding D-VNE* MULE Jointly embeds virtual nodes and links as much as possible Collapses multiple layers into one with information loss Collapses multiple layers into one without information loss * Zhang, et al. "Dynamic virtual network embedding over multilayer optical networks“, Journal of Optical Communications and Networking 7(9): 918-927, 2015.
  25. OPT-MULE: ILP model for optimal solution to MULE that minimizes

    bandwidth allocation cost for embedding VN and provisioning new IP links 28
  26. OPT-MULE* 29 Decision Variables Creation of new IP Links, IP

    Layer to Optical Layer Embedding for new IP links, Virtual Node and Link mapping Constraints  Typical VN Embedding constraints for VN to IP Layer Mapping  Newly created IP links must be embedded on the Optical layer  Port constraint for IP nodes  Capacity constraint for OTN links, * Details are in the paper
  27. OPT-MULE* 30 Decision Variables Creation of new IP Links, IP

    Layer to Optical Layer Embedding for new IP links, Virtual Node and Link mapping Constraints  Typical VN Embedding constraints for VN to IP Layer Mapping  Newly created IP links must be embedded on the Optical layer  Port constraint for IP nodes  Capacity constraint for OTN links, * Details are in the paper
  28. OPT-MULE* 31 Decision Variables Creation of new IP Links, IP

    Layer to Optical Layer Embedding for new IP links, Virtual Node and Link mapping Constraints  Typical VN Embedding constraints for VN to IP Layer Mapping  Newly created IP links must be embedded on the Optical layer  Port constraint for IP nodes  Capacity constraint for OTN links, * Details are in the paper
  29. OPT-MULE* 32 Decision Variables Creation of new IP Links, IP

    Layer to Optical Layer Embedding for new IP links, Virtual Node and Link mapping Constraints  Typical VN Embedding constraints for VN to IP Layer Mapping  Newly created IP links must be embedded on the Optical layer  Port constraint for IP nodes  Capacity constraint for OTN links, * Details are in the paper
  30. OPT-MULE* 33 Decision Variables Creation of new IP Links, IP

    Layer to Optical Layer Embedding for new IP links, Virtual Node and Link mapping Constraints  Virtual links can be mapped to existing or newly created IP links  Newly created IP links must be embedded on the Optical layer  Port constraint for IP nodes * Details are in the paper
  31. OPT-MULE* 34 Decision Variables Creation of new IP Links, IP

    Layer to Optical Layer Embedding for new IP links, Virtual Node and Link mapping Constraints  Virtual links can be mapped to existing or newly created IP links  Newly created IP links must be embedded on the Optical layer  Port constraint for IP nodes * Details are in the paper
  32. OPT-MULE* 35 Decision Variables Creation of new IP Links, IP

    Layer to Optical Layer Embedding for new IP links, Virtual Node and Link mapping Constraints  Virtual links can be mapped to existing or newly created IP links  Newly created IP links must be embedded on the Optical layer  Port constraint for IP nodes * Details are in the paper
  33. OPT-MULE* 36 Decision Variables Creation of new IP Links, IP

    Layer to Optical Layer Embedding for new IP links, Virtual Node and Link mapping Constraints  Virtual links can be mapped to existing or newly created IP links  Newly created IP links must be embedded on the Optical layer  Port constraint for IP nodes * Details are in the paper
  34. FAST-MULE: Challenges 39 Joint Embedding on IP and Optical Layer

    Challenge - I Solution Collapse IP and Optical Layer into a single layer
  35. FAST-MULE: Challenges 40 Joint embedding of virtual nodes and virtual

    links Joint Embedding on IP and Optical Layer Challenge - I Challenge - II Solution Collapse IP and Optical Layer into a single layer
  36. FAST-MULE: Challenges 41 Joint embedding of virtual nodes and virtual

    links Joint Embedding on IP and Optical Layer Challenge - I Challenge - II Solution Collapse IP and Optical Layer into a single layer Solution Embed star subgraphs from VN in a single shot using min-cost max-flow
  37. FAST-MULE: 3-Phase Algorithm 44 Phase-I (Collapse): Collapse IP and Optical

    Layers into a single layer collapsed graph Phase-II (Extract): Extract star subgraphs from VN
  38. FAST-MULE: 3-Phase Algorithm 45 Phase-I (Collapse): Collapse IP and Optical

    Layers into a single layer collapsed graph Phase-II (Extract): Extract star subgraphs from VN Phase-III (Embed): Jointly embed nodes and links of each star subgraph on the collapsed graph
  39. Phase-I: Collapse 46 A D E C B 1 3

    2 4 6 5 7 9 8 15 10 10 10 15 1000 985 990 990 990 985 1000
  40. Phase-I: Collapse 47 A D E C B 1 3

    2 4 6 5 7 9 8 1000 985 990 990 990 985 1000 Place as many direct links as the number of ports of an IP node to the corresponding OTN node (set bandwidth to port capacity) A 1 20 20 20 60 60 60 60 60
  41. Phase-I: Collapse 48 A D E C B 1 3

    2 4 6 5 7 9 8 10 15 1000 985 990 990 990 985 1000 10 10 15 • Place IP links between OTN nodes where the link’s IP endpoints are. • Keep IP link cost as is, set OTN link cost to very high. 60 60 60 60 60
  42. Phase-II: Extract 49 0 2 1 {C} {A, B} {D,

    E} 15 15 Extract star-shaped subgraph from VN Embedding a star-shaped subgraph in one-shot corresponds to jointly embedding a virtual node and all its incident virtual links. 15 0 2 1 {C} {A, B} {D, E} 15 15 2 1 {A, B} {D, E} 15
  43. 50 A D E C B 1 3 2 4

    6 5 7 9 8 10 15 1000 985 990 990 990 985 1000 10 10 15 0 Phase – III: Embed 2 1 {C} {A, B} {D, E} Star subgraph from VN Collapsed Graph 15 15 We reduce star-subgraph embedding to solving min-cost max-flow on collapsed graph. 60 60 60 60 60
  44. 51 A D E C B 1 3 2 4

    6 5 7 9 8 10 15 1000 985 990 990 990 985 1000 10 10 15 Phase – III: Embed 0 2 1 {C} {A, B} {D, E} 0 Map center node of star to one of its location constraint IP node. 15 15 60 60 60 60 60
  45. 52 A D E C B 1 3 2 4

    6 5 7 9 8 10 15 1000 985 990 990 990 985 1000 10 10 15 Phase-III: Embed 0   Add meta-node for each other Vnode. 0 2 1 {C} {A, B} {D, E} 15 15 60 60 60 60 60
  46. 53 A D E C B 1 3 2 4

    6 5 7 9 8 10 15 1000 985 990 990 990 985 1000 10 10 15 Phase-III: Embed 0   Add link from a VNode‘s location constraint nodes to its meta-node. 0 2 1 {C} {A, B} {D, E} 15 15 60 60 60 60 60
  47. Phase-III: Embed 54 A D E C B 1 3

    2 4 6 5 7 9 8 10 15 1000 985 990 990 990 985 1000 10 10 15  t  3 3 3 3 3 1 1 1 1 1 1 Add a sink node (t). Add unit capacity link from all meta-nodes to sink node. 0 2 1 {C} {A, B} {D, E} 0 15 15
  48. Phase-III: Embed 55 A D E C B 1 3

    2 4 6 5 7 9 8 0 1 66 65 65 65 65 65 66 0 0 1  t  3 3 3 3 3 1 1 1 1 1 1 0 0 2 1 {C} {A, B} {D, E} 15 15 Set cap. of other links to: max. number of VLinks that can be placed on that link
  49. Phase-III: Embed 56 A D E C B 1 3

    2 4 6 5 7 9 8 0 1 66 65 65 65 65 65 66 0 0 1  t  3 3 3 3 3 1 1 1 1 1 1 0 0 2 1 {C} {A, B} {D, E} 1 2 15 15 Solve min-cost max-flow to obtain joint node and link embedding.
  50. Evaluation: Setup  FAST-MULE compared with OPT-MULE and D-VNE* 

    OTN 15 – 100 nodes  IP Network ~60% the size of the OTN Virtual Network  4 – 8 nodes  20 VNs for each IP/OTN combination 57 * Zhang, et al. "Dynamic virtual network embedding over multilayer optical networks“, Journal of Optical Communications and Networking 7(9): 918-927, 2015.
  51. FAST-MULE Performance Highlights 59 67% better than D-VNE on avg.

    Optimal for star shaped VN* * Proof is in the paper
  52. FAST-MULE Performance Highlights 60 Within ~47% of optimal on avg.

    67% better than D-VNE on avg. Optimal for star shaped VN* * Proof is in the paper
  53. FAST-MULE Performance Highlights 61 Within ~47% of optimal on avg.

    2-3 Orders of magnitude faster than OPT-MULE 67% better than D-VNE on avg. Optimal for star shaped VN* * Proof is in the paper
  54. Summary 62 We address VNE problem for Multi-Layer IP-over- OTN

    Network Two Solutions to MULE: OPT-MULE, FAST-MULE FAST-MULE performs ~47% better than the optimal (empirically); allocates ~66% less resources than the state-of-the-art
  55. What’s Next? 63 Can we exploit topological flexibility for failure

    recovery? What is the impact of fragmentation? How challenging is it to address MULE for other Optical network technologies (e.g., Elastic Optical Networks)?
  56. 64

  57. FAST-MULE: Complexity 66 ( ′ 2 log ) V’ =

    Number of Virtual nodes V = Number of nodes in collapsed graph E = Number of links in collapsed graph
  58. Conflict Resolution using “Referee Node” 67 A D E C

    B 1 3 2 4 6 5 7 9 8 0 1 66 65 65 65 65 65 66 0 0 1  t  3 3 3 3 3 1 1 1 1 1 0 0 2 1 {C} {A, B} {A, E} 15 15 Add meta link: Conflicting nodereferee nodemeta-node r 1 1 1