ReViNE: Reallocation of Virtual Network Embedding to Eliminate Substrate Bottlenecks

Perceived as a key enabling technology for the future Internet, Network Virtualization (NV) allows an Infrastructure Provider (InP) to better utilize their Substrate Network (SN) by provisioning multiple Virtual Networks (VNs) from different Service Providers (SPs). A key challenge in NV is to efficiently map the VN requests from SPs on an SN, known as the Virtual Network Embedding (VNE) problem. VNE algorithms are typically online in nature. A VN embedding can become suboptimal over time due to the arrival and departure of other VNs as well as due to changes in SN such as failures. One way to mitigate the impact of such dynamism is to periodically reallocate resources for the existing VNs. VNE reallocation can increase an InP’s revenue by decreasing bandwidth consumption and by increasing the possibility of accepting future VNs. In this paper, we study Reallocation of Virtual Network Embedding (ReViNE) problem to minimize the number of over utilized substrate links and total bandwidth cost on the SN. We propose an Integer Linear
Programming formulation for the optimal solution (ReViNEOPT) and a simulated annealing based heuristic (ReViNE-FAST) to solve larger problem instances. Simulation results show that on average our proposed heuristic performs within ~19% of the optimal solution. Moreover, ReViNE-FAST generates more than 2.5× better solutions compared to the state-of-the-art simulated annealing based heuristic for VNE reallocation.

Transcript

1. ReViNE: Reallocation of Virtual Network Embedding to Eliminate Substrate Bottleneck Shihabur R. Chowdhury, Reaz Ahmed, Nashid Shahriar, Aimal Khan, 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 60 100 55 50 70 85 85 110 27 15 12 10 20 22 22 20 25 a b c
4. Impact of Dynamicity: An Empirical Study 4 300+ SNodes, 900+ SLinks. (AS6461), 4 – 8 VNodes/VN (50% conn. pr.) Poisson Arrival (10VNs/100 T.U.), Exponential Lifetime (1000 T.U.) Optimal embedding that minimizes total bandwidth consumption
5. Impact of Dynamicity: An Empirical Study 5 Acceptance Ratio capped at ~50% 300+ SNodes, 900+ SLinks. (AS6461), 4 – 8 VNodes/VN (50% conn. pr.) Poisson Arrival (10VNs/100 T.U.), Exponential Lifetime (1000 T.U.) Optimal embedding that minimizes total bandwidth consumption
6. Impact of Dynamicity: An Empirical Study 6 ~30% links utilized >= 70% ~40% links utilized <= 10%
7. Impact of Dynamicity: An Empirical Study 7 ~30% links utilized >= 70% ~40% links utilized <= 10% Skewed Substrate Link Utilization impacts Acceptance Ratio !!
8. Key Question: How to cope with the dynamicity in Network Virtualization when little or no information about the future is available? 8
9. (One Possible) Answer: Periodically adjust the embedding to eliminate “bottlenecks” and “optimize resource usage” 9
10. The Problem Reallocation of Virtual Network Embedding (ReViNE) 10 Given a Substrate Network and a set of embedded Virtual Networks
11. The Problem Reallocation of Virtual Network Embedding (ReViNE) 11 Migrate Virtual Nodes to New Substrate Nodes Given a Substrate Network and a set of embedded Virtual Networks
12. The Problem Reallocation of Virtual Network Embedding (ReViNE) 12 Migrate Virtual Nodes to New Substrate Nodes Migrate Virtual Links to New Substrate Paths Given a Substrate Network and a set of embedded Virtual Networks
13. The Problem Reallocation of Virtual Network Embedding (ReViNE) 13 Migrate Virtual Nodes to New Substrate Nodes Migrate Virtual Links to New Substrate Paths Objective: Eliminate Substrate Bottlenecks* and Minimize Resource Usage** Given a Substrate Network and a set of embedded Virtual Networks * Links with utilization >= % ** In our case, bandwidth consumed by virtual links
14. Our Proposal 14 ReViNE-OPT ReViNE-FAST ILP-based optimal solution* (NP-Hard) Simulated Annealing-based heuristic A suit of solutions to ReViNE * Details is in the paper
15. Do We Need A Heuristic? 15 Computing Optimal Solution is Very Expensive H/W Configuration: 8x10 Core Intel Xeon E5 CPU, 1TB RAM Observed limits for ILP: 50 – 100 Node SN with < 60VNs took several hours and several 10s of GB RAM ILP Can Yield Impractical Solutions  A practical solution contains a sequence of operations to reach the re-optimized state (also satisfy make-before-break constraint)  Not possible to model in ILP. Final state obtained from ILP can be unreachable without violating make-before-break constraint.
16. Do We Need A Heuristic? 16 Computing Optimal Solution is Very Expensive H/W Configuration: 8x10 Core Intel Xeon E5 CPU, 1TB RAM Observed limits for ILP: 50 – 100 Node SN with < 60VNs took several hours and several 10s of GB RAM ILP Can Yield Impractical Solutions  A practical solution contains a sequence of operations to reach the re-optimized state (also satisfy make-before-break constraint)  Not possible to model in ILP. Final state obtained from ILP can be unreachable without violating make-before-break constraint.
17. Do We Need A Heuristic? 17 Computing Optimal Solution is Very Expensive H/W Configuration: 8x10 Core Intel Xeon E5 CPU, 1TB RAM Observed limits: 50 – 100 Node SN with < 60VNs took several hours and several 10s of GB RAM ILP Can Yield Impractical Solutions  A practical solution contains a sequence of operations to reach the re-optimized state (also satisfy make-before-break constraint)  Hard to model in ILP. Final state obtained from ILP can be unreachable without violating make-before-break constraint.
18. Heuristic Design 18 Our Objectives are Conflicting Minimize Bottleneck Links Minimize Bandwidth Usage
19. Heuristic Design 19 Our Objectives are Conflicting Minimize Bottleneck Links Minimize Bandwidth Usage Distribute load across substrate links Paths can become longer
20. Heuristic Design 20 Our Objectives are Conflicting Minimize Bottleneck Links Minimize Bandwidth Usage Distribute load across substrate links Route Virtual Links on Shorter Paths Substrate links on shorter paths can become bottlenecks Paths can become longer
21. Heuristic Design 21 Our Objectives are Conflicting Minimize Bottleneck Links Minimize Bandwidth Usage Distribute load across substrate links Route Virtual Links on Shorter Paths Substrate links on shorter paths can become bottlenecks Paths can become longer
22. Heuristic Design 22 Our Objectives are Conflicting Minimize Bottleneck Links Minimize Bandwidth Usage Distribute load across substrate links Route Virtual Links on Shorter Paths Substrate links on shorter paths can become bottlenecks Paths can become longer Instead of an one-shot algorithm, use a meta-heuristic (Simulated Annealing) to explore the solution space and find a balance.
23. Simulated Annealing: Neighborhood Generation 23 Randomly select a VN and reroute a randomly selected virtual link. Bottleneck Substrate Link Reconfiguration Select a bottleneck substrate link and reroute virtual links using that bottleneck link until it is no longer a bottleneck. Virtual Node Migration Randomly select a VN and re-embed a random virtual node and incident virtual links. Virtual Link Migration
24. Simulated Annealing: Neighborhood Generation 24 Randomly select a VN and reroute a randomly selected virtual link. Bottleneck Substrate Link Reconfiguration Select a bottleneck substrate link and reroute virtual links using that bottleneck link until it is no longer a bottleneck. Virtual Node Migration Randomly select a VN and re-embed a random virtual node and incident virtual links. Virtual Link Migration
25. Simulated Annealing: Neighborhood Generation 25 Randomly select a VN and reroute a randomly selected virtual link. Bottleneck Substrate Link Reconfiguration Select a bottleneck substrate link and reroute virtual links using that bottleneck link until it is no longer a bottleneck. Virtual Node Migration Randomly select a VN and re-embed a random virtual node and incident virtual links. Virtual Link Migration
26. Exploiting Multi-core CPU 26 Parallel Simulated Annealing Searches Initial State Seed Solution-0 Seed Solution-1 Seed Solution-k
27. Exploiting Multi-core CPU 27 Parallel Simulated Annealing Searches Initial State Seed Solution-0 Seed Solution-1 Seed Solution-k Search Thread – 0 (CPU0) Search Thread – 1 (CPU1) Search Thread – k (CPUk)
28. Exploiting Multi-core CPU 28 Parallel Simulated Annealing Searches Initial State Seed Solution-0 Seed Solution-1 Seed Solution-k Search Thread – 0 (CPU0) Search Thread – 1 (CPU1) Search Thread – k (CPUk) Best Solution from All Searches
29. Evaluation: Setup  ReViNE-FAST compared with ReViNE-OPT and SA-realloc*  Parameters  50 – 100 node synthetic substrate network  Larger test cases with 1000 node (heuristic only comparison)  Mean degree between 3.6 – 4  Mean substrate link utilization 60% - 80%  Bottleneck substrate link threshold 70% - 90% 29 * Masti, S,. et al. “Simulated Annealing Algorithm for Virtual Network Reconfiguration“, 8th Euro-NGI Conference on Next Generation Internet, IEEE, 2012, pp. 95-102.
30. ReViNE-FAST Performance Highlights 30 Within ~19% of optimal (ReViNE-OPT) on avg. ~3x less cost compared to SA-realloc on avg. ~5% more VNs accepted on avg. when combined with optimal VN embedding algorithm
31. Summary 31 ReViNE is one possible way to address the dynamicity in VN arrival/departure ReViNE-FAST, a simulated annealing based heuristic performs ~19% within the optimal (empirically evaluated) ReViNE-FAST performs ~3x better than S.O.A Simulated Annealing-based heuristic
32. Questions? 32 Source Code CPLEX: https://github.com/srcvirus/vne-reallocation-cplex Simulated Annealing: https://github.com/srcvirus/vne-reallocation-sa
33. 33 Backup
34. ReViNE-FAST vs ReViNE-OPT 34
35. Impact of Reallocation 35
36. ReViNE-FAST vs SA-Realloc (Large Cases) 36
37. ReViNE-FAST Convergence (Large Cases) 37
38. State-of-the-art 38 Proactive One-shot Approaches: Periodically reallocate VNs [3][4] Reactive One-shot Approaches: Reallocate VNs when a new VN cannot be embedded [1][2] Meta-heuristic Approaches: Simulated Annealing [5], Particle Swarm Optimization [6] [1] Y. Zhu et al., “Algorithms for assigning substrate network resources to virtual network components”, IEEE INFOCOM, 2006. [2] M. Yu, et al. “Rethinking virtual network embedding: substrate support for path splitting and migration”, ACM SIGCOMM CCR, 38(2), 2008, pp. 17–29. [3] N. F. Butt, et al. “Topology-awareness and reoptimization mechanism for virtual network embedding”, Int. Conf. on Research in Networking 2010. [4] P. N.Tran, et al., “Optimal mapping of virtual networks considering reactive reconfiguration,” IEEE CloudNet, 2012. [5] S. Masti, et al. “Simulated Annealing Algorithm for Virtual Network Reconfiguration“, 8th Euro-NGI Conf. on Next Generation Internet, IEEE, 2012. [6] Y. Yuan, et al. ,“Discrete particle swarm optimization algorithm for virtual network reconfiguration,” Int. Conf. in Swarm Intelligence, 2013.