1. E E 681 - Module 10 © Wayne D. Grover 2002, 2003 1 Introductory Briefing on RingBuilder™ Research prototype ring-network design system developed 1997-2001 at TRLabs by W. D. Grover, D. Morley, (PhD candidate) J. Slevinsky, (Telus Industrial rep., MSc candidate) M. Jeremiah (U of A coop student) J. Hopkins, Nortel (lead user / advisor) W. D. Grover TRLabs & University of Alberta © Wayne D. Grover 2002, 2003 E E 681 - Module 10

2. E E 681 - Module 10 © Wayne D. Grover 2002, 2003 2 Multi-Ring Network Design Problem (RingBuilder view) RingBuilder (or any other design method ) Given Network topology Demand pattern Ring technologies Cost models ~ Min-cost Design Ring System decisions  Type  OC-n size  Topological layout  Glass-through locations Routing plan  Ring assignment  Inter-ring transit locations Subject to: All demands served Ring capacity constraints Max. ADMs per ring Limited Inter-ring transit locations Partial add/drop constraints ( Matched-nodes requirements, etc.)

3. E E 681 - Module 10 © Wayne D. Grover 2002, 2003 3 RingBuilder.... RingBuilder has a high degree of “fidelity” (realism in modeling the actual problem in all its details) but is a sub-optimal (heuristic-based) design system. Output designs are fully specified, feasible to construct directly. Based on a central “greedy” hypothesis: - that a good network design is comprised of good individual rings - that a complete network design can be developed by choosing “good” rings one after another until all demands are served. As RingBuilder developed, it includes an increasing number of tactics to “overcome” this greediness in terms of solution quality, while retaining the basic iterative - synthesis framework.

4. E E 681 - Module 10 © Wayne D. Grover 2002, 2003 4 RingBuilder.... (ver. 3) Basic Method: - Phase 1 pre-processing steps (i) Candidate Generation: Depth-first search algorithm used to enumerate all distinct simple cycles. (ii) Demand Routing: Route point-to-point demands via shortest path over basic graph topology. Split flows over equal shortest routes if they exist. - Phase 2 Iterative Design synthesis (i) Solve loading problem for each distinct cycle in each ring technology in current environment of un-served demand segments. (optimal or heuristic). (ii) Choose and place ring candidate with highest measure of transport utility, (iii) Demand packing: exploit the rings just (or so far) placed to convey any un-served demand segments. (iv) Update un-served segments of remaining routes. - Phase 3 (optional) Improvement heuristics or repeat Phase 2 with random variation

5. E E 681 - Module 10 © Wayne D. Grover 2002, 2003 5 Phase 1: Candidate Generation –Each combination of network cycle and ring technology (i.e., ring type and line rate) represents a ring candidate. RingBuilder....Design Methodolgy Example:

6. E E 681 - Module 10 © Wayne D. Grover 2002, 2003 6 Phase 1: Initial Demand Routing –Point-to-point demands are routed across the network topology before any rings are placed using shortest path algorithm (min-hop or min-distance). RingBuilder....Design Methodolgy Example: 1 42 5 3 6 Min-hop routingPossible min-distance (cost) routing may differ 1 42 5 3 6 Point-to-point demand matrix need: algorithms for shortest path routing (Dijkstra)

7. E E 681 - Module 10 © Wayne D. Grover 2002, 2003 7 Phase 2: Candidate Ring Loading (heuristic method) –Where the route of a demand flow intersects a ring candidate the relevant demand segments are loaded onto each candidate ring in decreasing order of length served (or capture achieved) until all segments are loaded or capacity is exhausted. RingBuilder....Design Methodolgy Example (by segment length priority): Working capacity Spare (unused) Protection capacity ADM Glass-through Node Self-healing Ring 2nd demand loaded 3rd demand loaded 4th & 5th demands loaded 1st demand loaded

8. E E 681 - Module 10 © Wayne D. Grover 2002, 2003 8 Phase 3: Ring Selection (and placement). RingBuilder....Design Methodolgy max feasible loading total detailed cost of constructing the corresponding ring of type k on cycle j. i.e., Note that choice of demand segments in numerator, determines where the glass-through nodes are and all low-speed add / drop circuit pack costs that are incurred

9. E E 681 - Module 10 © Wayne D. Grover 2002, 2003 9 RingBuilder....Simple example (1) Point-to-point demand matrix 1 32 456 Network Topology 1 32 456 52 11 17 44 11 12 Min-Hop Demand Routing This example based on assuming 4 fiber OC-12 or 2 fiber OC-24 rings Ring selection based on: revised Oct 24, 2000

10. E E 681 - Module 10 © Wayne D. Grover 2002, 2003 10 RingBuilder....Simple example (2) 1 32 456 1 32 456 1 32 456 1 32 456 1 32 456 1 32 456 1 32 456 Cycle Finding 1 32 456 1 32 456 1 32 456 1 32 456 1 32 456 1 32 456 Ring Loading 1st Iteration 1 32 456 52 11 17 44 11 12 *Assume BLSR/4 OC-12 Ring #1 revised Oct 24, 2000

11. E E 681 - Module 10 © Wayne D. Grover 2002, 2003 11 RingBuilder....Simple example (3) Remove covered route segments 1 32 456 11/12 12/12 4/12 11/12 12/12 1 32 456 52 0 5 40 0 0 Remaining un-served demand segments Ring #1 1 32 456 1 32 456 1 32 456 1 32 456 1 32 456 1 32 456 1 32 456 52 0 5 40 0 0 Ring #2 Ring Loading 2nd Iteration revised Oct 24, 2000

12. E E 681 - Module 10 © Wayne D. Grover 2002, 2003 12 RingBuilder....Simple example (4) Remove covered route segments 1 32 456 0/0 5/12 1 32 456 52 0 0 00 0 0 Remaining un-served demand segments Ring #2 1 32 456 1 32 456 1 32 456 1 32 456 1 32 456 1 32 456 1 32 456 52 0 0 00 0 0 4/12 Ring Loading 3rd Iteration Ring #3 revised Oct 24, 2000

13. E E 681 - Module 10 © Wayne D. Grover 2002, 2003 13 RingBuilder....Simple example (5) Remove covered route segments 1 32 456 5/12 2/12 1 32 456 00 0 0 00 0 0 Remaining un-served demand segments Ring #3 0/12 1 32 456 11/11 12/12 4/12 11/12 12/12 1 32 456 5/12 2/12 0/12 1 32 456 5/12 4/12 Final Ring-cover Network Design (and resultant loadings)

14. E E 681 - Module 10 © Wayne D. Grover 2002, 2003 14 RingBuilder.... Main User Interface

15. E E 681 - Module 10 © Wayne D. Grover 2002, 2003 15 RingBuilder.... “Advisor” Mode

16. E E 681 - Module 10 © Wayne D. Grover 2002, 2003 16 Summary: State of the art and Research Directions in Multi-Ring Network Design Solution Quality Model Accuracy Eulerian Ring Covers (Gardner et al., ‘94). Ring Coverage IP (Kennington, ‘97). RingBuilder™ (Slevinsky,Grover, ‘93) Net-Solver (Gardner et al., ‘95) Simulated Annealing (Roberts, ‘94). Hierarchical Rings (Shi,Fonseka, ‘96). Strategic Options (Wasem,Wu ‘91) Research Goals RingBuilder™ (Slevinsky,Grover, ‘95) CapacitatedMulti-technologyMulti-period Probabilistic Topology

17. E E 681 - Module 10 © Wayne D. Grover 2002, 2003 17 Other approaches to Multi-Ring Network Design Preliminary: Concept of “ideal” or “topological” rings An idealized or purely topological ring has architectural properties or other figures of merit relative to some problem, but is not modular like a real ring and in fact has no assumed capacity limit. Examples: idealized ring may still have attributes of: - spans covered - working capacity balance - capture efficiency - total mileage Purpose is usually to permit problem simplification while still identifying high merit (low redundancy) ring-layouts or efficient span cover solutions, etc.

18. E E 681 - Module 10 © Wayne D. Grover 2002, 2003 18 Other approaches: RingBuilder versions 1, 2 ver 1. sought to minimize redundancy of an ideal ring span cover of the graph demands are shortest-path routed beforehand, cycle set enumerated beforehand greedy-iterative buildup of a ring cover based on choosing ring with best balance efficiency at each iteration. ver. 2, also assessed each ring candidate for demand capture efficiency - balance and capture efficiency measures were ‘blended’ into a single combined figure of merit. - expressions for balance & capture efficiency are those already given. - ver 2. also extended into modular ring case: - loading algorithms would select demands to try to maximize balance or capture - hypothesis was that a cost-optimal design had to represent the best trade-off of capture versus balance effects. - hence tactic of a “sweep” of alpha from 0 to 1 to identify a near min cost design - select rings on bal-capture merit, but measure design by detailed costing including glass-through details etc.

19. E E 681 - Module 10 © Wayne D. Grover 2002, 2003 19 Other approaches: results show hypothesis is basically sound, but overall method had following drawbacks : - inherently a “single technology” design approach - alpha sweep onerous, and only a surrogate-based search for cost minimum RingBuilder version 2 actual design cost low alpha min-cost for metro high alpha min-cost for metro

20. E E 681 - Module 10 © Wayne D. Grover 2002, 2003 20 Pure “Balance-optimized” Design  = 1.0 Total transitions = 2970 Avg. balance efficiency = 0.47 Example from RingBuilder 2: 9 rings

21. E E 681 - Module 10 © Wayne D. Grover 2002, 2003 21  = 0.0 Total transitions = 1698 Avg. balance efficiency = 0.32 Pure “Capture-optimized” Design Example from RingBuilder 2: 11 rings

22. E E 681 - Module 10 © Wayne D. Grover 2002, 2003 22  = 0.4 Total transitions = 1918 Avg. balance efficiency = 0.44 Compromise Design near min cost found empirically at  ~0.4 Example from RingBuilder 2: fewest rings of all: 8 rings