Survivor: an Enhanced Controller Placement Strategy for Improving - PowerPoint PPT Presentation

Survivor: an Enhanced Controller Placement Strategy for Improving SDN Survivability Lucas F. Müller , Rodrigo R. Oliveira, Marcelo C. Luizelli, Luciano P. Gaspary, Marinho P. Barcellos Federal University of Rio Grande do Sul (UFRGS), Brazil 57th IEEE Global Communications Conference (GLOBECOM 2014) December 8 - 12, 2014 Austin, Texas – EUA

Software-Defined Networking Design • Changing the way networks are designed and managed • Separates the control plane from the data plane • Moves the control logic to an external entity (Controller) • Controller provides resources and abstractions to facilitate programming … Despite its benefits, SDN created an inherent dependency relationship between forwarding devices and the controller. Lucas F. Müller Survivor: Controller Placement Survivability – 2

Software-Defined Networking Design Controller Placement Problem Set of SDN Controllers Forwarding devices Lucas F. Müller Survivor: Controller Placement Survivability – 3

Software-Defined Networking Design Controller Placement Problem Lucas F. Müller Survivor: Controller Placement Survivability – 4

Software-Defined Networking Design Controller Placement Problem Lucas F. Müller Survivor: Controller Placement Survivability – 5

Software-Defined Networking Design Controller Placement Problem Lucas F. Müller Survivor: Controller Placement Survivability – 6

Software-Defined Networking Design Controller Placement Problem Lucas F. Müller Survivor: Controller Placement Survivability – 7

Software-Defined Networking Design Controller Placement Problem Lucas F. Müller Survivor: Controller Placement Survivability – 8

Software-Defined Networking Design Controller Placement Problem 0 0 0 0 0 Lucas F. Müller Survivor: Controller Placement Survivability – 9

Software-Defined Networking Design Controller Placement Problem 0 0 0 0 0 1 1 1 1 Lucas F. Müller Survivor: Controller Placement Survivability – 10

Software-Defined Networking Design Controller Placement Problem 0 0 0 0 2 2 0 1 1 2 1 1 2 Lucas F. Müller Survivor: Controller Placement Survivability – 11

Software-Defined Networking Design Controller Placement Problem 0 0 0 0 2 2 0 1 3 1 2 3 1 3 1 3 2 Lucas F. Müller Survivor: Controller Placement Survivability – 12

Software-Defined Networking Design Controller Placement Problem 0 4 4 4 0 0 4 4 4 0 4 2 4 4 2 0 1 3 1 2 3 1 3 1 3 2 Ok, the control plane design is ready. Lucas F. Müller Survivor: Controller Placement Survivability – 13

“The network is down.” Lucas F. Müller Survivor: Controller Placement Survivability – 14

Software-Defined Networking Design 0 0 4 ? 4 ? 0 0 4 0 0 ? X X 4 4 4 ? 4 ? 0 0 2 4 2 ? 4 ? 4 4 0 2 0 2 1 1 3 ? 1 X 1 2 2 3 3 1 1 3 3 1 3 1 3 2 2 Single link failures Multiple connectivity failures 0 4 4 0 0 4 4 # 9 # 5 4 4 # 4 0 2 4 4 4 0 2 1 # 4 3 1 2 3 1 Controller overload 3 # 4 1 3 2 Lucas F. Müller Survivor: Controller Placement Survivability – 15

Controller Placement Strategy for Improving SDN Survivability Goal : novel controller placement strategy that deals with control plane survivability in large scale SDN networks. Provide and maintain network services in face of operational challenges React and attempt to recover from harmful events Lucas F. Müller Survivor: Controller Placement Survivability – 16

Outline • Introduction : context and motivation • Proposed Approach : strategy and modeling • Results : resilience and overload • Conclusion Lucas F. Müller Survivor: Controller Placement Survivability – 17

Proposed Approach Goals – Connectivity Increase path diversity between device-controller – Capacity Avoid controller overload – Recovery Define a methodology for composing smarter failover mechanisms Lucas F. Müller Survivor: Controller Placement Survivability – 18

Proposed Approach: Overview Divided in two complementary parts – Defines the placement of controllers instances – Compose the list of backup controllers for each device in the network Lucas F. Müller Survivor: Controller Placement Survivability – 19

Proposed Approach: two complementary parts – Defines placement for controller instances 32 3 37 0 35 0 36 0 38 3 39 5 2 0 31 3 33 3 30 2 34 0 2 0 19 5 3 5 1 0 10 5 4 5 2 2 6 5 7 2 4 23 29 2 11 1 8 0 22 1 28 27 4 1 9 4 13 1 26 1 24 25 3 2 12 1 21 1 20 4 4 14 15 4 5 16 18 4 17 5 Controller placement Network topology Lucas F. Müller Survivor: Controller Placement Survivability – 20

Proposed Approach: two complementary parts – Specifies backup controllers for each device in the network 2 0 3 4 4 0 0 5 1 0 3 5 0 1 4 2 3 3 2 2 0 5 0 5 2 5 0 5 2 2 5 2 4 1 2 1 0 1 4 1 4 1 1 0 3 2 1 1 4 4 3 4 5 4 5 Lucas F. Müller Survivor: Controller Placement Survivability – 21

Proposed Approach: modeling Optimal Linear Model for Controller Placement – Strategy modeled as optimization problem – Achieve the optimal solution – Survivor strategy: Integer Linear Program, 1 objective (maximize connectivity between device-controller) Heuristics for Defining Lists of Backup Controllers – Compose the lists of backup controllers – Eliminating the need to manually determine the list – Proximity and Residual capacity-based heuristics – Proposed generic framework for designing heuristics Lucas F. Müller Survivor: Controller Placement Survivability – 22

Outline • Introduction : context and motivation • Proposed Approach : strategy and modeling • Results : resilience and overload • Conclusion Lucas F. Müller Survivor: Controller Placement Survivability – 23

Methodology Configuration – Three different WAN topologies: Internet2 (10 nodes, 15 links), RNP (27 nodes, 33 links) and GÉANT (40 nodes, 61 links) – Controllers capacity: 1800 kilorequests/s – Forwarding devices requests: 200 kilorequests/s – Percentage of controller backup resources: 30% Comparison method – Resilient placement strategy Zhang et al., denoted by MCC [CUNHA et al., 2009; KNIGHT et al., 2011; TOOTOONCHIAN et al., 2012; ZHANG et al., 2011] Lucas F. Müller Survivor: Controller Placement Survivability – 24

Methodology Four metrics – Resilience • Resilience equation used by Zhang et al., 2011 • Cardinal of edge-connectivity – Overload • Number of overloaded controllers • Load distribution for each of the controller instances [CUNHA et al., 2009; KNIGHT et al., 2011; TOOTOONCHIAN et al., 2012; ZHANG et al., 2011] Lucas F. Müller Survivor: Controller Placement Survivability – 25

Results: resilience Probability of connectivity loss (Resilience equation, Zhang et. al) SVVR MCC SVVR MCC GEANT RNP INTERNET2 GEANT RNP INTERNET2 0.5 1 Probability of connectivity loss Probability of connectivity loss 0.4 0.8 gain 0.3 0.6 gain 0.2 0.4 0.1 0.2 0 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1 0 0.2 0.4 0.6 0.8 1 Failure Probability Failure Probability (a) 1% a 10% (b) 0% a 100% Survivor reduces the probability of connectivity loss. Lucas F. Müller Survivor: Controller Placement Survivability – 26

Results: resilience Effect of exploring path diversity (Cardinal of edge-connectivity) SVVR MCC SVVR MCC+ 1 1 1 1 3 0.8 3 0.8 % of failure scenarios % of failure scenarios 6 0.6 0.6 6 0.4 0.4 0.2 0.2 0 0 0 5 10 15 20 0 2 4 6 8 10 # of Disconnected elements # of Disconnected elements CDFs of disconnected devices for all possible cases of 1, 3 and 6 link disruptions Path diversity increases the network survivability, and it requires explicit consideration to be fully explored. Lucas F. Müller Survivor: Controller Placement Survivability – 27

Results: overload Number of overload scenarios 20 Normal Failover strategy: Failover strategy: # of overload scenarios Operation Proximity Residual Capacity 15 10 5 0 SVVR MCC SVVR MCC SVVR MCC Network convergence after disruptions is highly sensible to predefined information in failover mechanisms. Lucas F. Müller Survivor: Controller Placement Survivability – 28

Results: overload Network state after convergence (Load distribution) SVVR MCC SVVR MCC 300 300 250 250 200 200 Load (%) Load (%) 150 150 100 100 50 50 0 0 C1 C2 C3 C4 C5 C6 C7 C1 C2 C3 C4 C5 C6 C7 C1 C2 C3 C4 C5 C6 C7 C1 C2 C3 C4 C5 C6 C7 (a) Proximity heuristic (b) Residual Capacity heuristic Controller overload can be handled proactively by adding capacity-awareness and setting backup resources. Lucas F. Müller Survivor: Controller Placement Survivability – 29

Outline • Introduction : context and motivation • Proposed Approach : strategy and modeling • Results : resilience and overload • Conclusion Lucas F. Müller Survivor: Controller Placement Survivability – 30

Final Remarks Contributions – Significant reduction on connectivity loss – More realistic controller placement strategy – Smarter recovery mechanisms – Optimization model in order to generate optimal results Ongoing work – Studying meta-heuristics – Extend evaluation Lucas F. Müller Survivor: Controller Placement Survivability – 31