Globecom 2003 December 1-5, San Francisco, California On the Efficacy of GMPLS Auto-Reprovisioning as a Mesh-Network Restoration Mechanism Govind V. Kaigala & Wayne D. Grover TRLabs and University of Alberta Edmonton, AB, Canada Globecom 2003 December 1-5, San Francisco, California Group URL : www.ece.ualberta.ca/~grover/
Outline GMPLS Auto-reprovisioning and motivation “Mutual capacity” and Multi-Commodity Max-Flow (MCMF) Experimental methods and results Effect of excess capacity Recommendations
Background and Motivation Rings Span-restorable Mesh Path-restorable Mesh p-Cycles Shared Backup Path Protection (SBPP) KNOWN RESTORATION AND PROTECTION SCHEMES GMPLS Mass-redial
GMPLS “Mass-redial” Message-handling behavior of nodes and protocols. Deleterious interactions between simultaneous CR-LDP instances. “Mass-redial” - Reattempts compound dynamic deadlocks, fall backs and re-attempts. (b) “mutual-capacity” allocation issues.
Example to Illustrate mutual coordination issue Span failure 3 O-D pairs affected (1 unit each) A C 1 B D E 3 service paths are lost
Example to Illustrate mutual coordination issue Span failure 3 O-D pairs affected (1 unit each) 2 possible sequences Sequence 1 A Result R=2/3 1 1 1 Reserve capacities 1 1 C 1 1 Sequence of seizure B 1 3 service paths are lost 1 1 1 1 D Route computation 1 1 E 1 1 Routing : Cost metric
Example to Illustrate mutual coordination issue Span failure 3 O-D pairs affected 2 possible sequences Sequence 1 Sequence 2 A Result R=2/3 Result R=1/3 1 1 1 Reserve capacities 1 1 C 1 1 B 1 1 1 1 1 D 1 1 E 1 1 Routing : Cost metric
Concept of “Mutual capacity” - Coordination True Path Restoration poses a form of a capacitated multi-commodity flow problem. How much of each flow to route over each edge coupled with that for every other commodity by the edge capacity constraints. MCMF Finite capacities on edges Shortest path (least cost) routing for each commodity is not optimal.
Experimental Methodology Define minimal-capacity test networks. Validate 100% restorability with true path restoration. For each failure scenario: Simulate GMPLS mass-redial process. Generate unbiased random samples for all affective sequences. (d) Measure average and individual node-pair restoration levels.
Test Networks Dijkstra’s shortest path algorithm is used to find the route for a working path. Demand matrixes are randomly generated within the range [1-20] for each node pair. Adequate wavelength conversion at the OXC nodes. Net-2 Net-3 IXC-Net-1
Experimental Results (1) Sample data showing sequence-dependent variation in restorability. Net-3
Experimental Results (2) Star plot view of pair-wise restorability. Net-3 Expected Scenario-1 Scenario-2 Average pair-wise restorability Stochastic outcomes
Experimental Results (3) Pair-wise restorability Statistics. Worst Case Situations… X, Ri over all trials Net-1(IXC) Net-2 Net-3 < 30% 10% 31% 18% < 50% 14.3% 39% 26% Sample numbers and variance considerations indicate 95% confidence intervals are all under 6.3% in estimating the average restorability level.
Effect of Excess Capacity Average Over-provisioning of 160-175% Spare Capacity to: mitigate the unpredictability. mitigate the worst case outcomes.
Experimental Results (4) Effect of Nodal Degree ++ Sparse Network -- Dense Network Why? Performance is poorer in highly connected networks. Spare network tends to have less salability and less dependence on precise mutual capacity coordination. In the limit of sparseness, a ring is reached for which the mutual capacity issue does not exists.
Summary and Recommendations GMPLS Auto-reprovisioning cannot give precise or repeatable performance – No definite SLA. Repeatable performance – only by lavish over-provisioning. GMPLS re-dial by itself is not a recommended restoration mechanism for basic transport networks. Preferred scheme - guarantee, by design, on the basis of capacity design. Isolated Path Failure X Cable Cut
Globecom 2003 December 1-5, San Francisco, California On the Efficacy of GMPLS Auto-Reprovisioning as a Mesh-Network Restoration Mechanism Govindkrishna V. Kaigala & Wayne D. Grover TRLabs and University of Alberta Edmonton, AB, Canada www.ece.ualberta.ca/~grover/ Globecom 2003 December 1-5, San Francisco, California