1. A New Approach to Node-Failure Protection with Span-Protecting p-Cycles
Presented by: Wayne D. Grover,
(co-author with Diane Prisca Onguetou)
TRLabs and the University of Alberta, Canada
11th International Conference on Transparent Optical Networks / 5th Reliability Issues in Next Generation Networks (ICTON/RONEXT)
June 30th 2009, Island of Sao Miguel, Azores, Portugal
IEEE ICTON/RONEXT (June 30th 2009), Wayne D. Grover and Diane P. Onguetou,
Node Failure Recovery with Ordinary p-Cycles.

2. OUTLINE Background and Objectives Current Approaches
The Novel Strategy
Experimental Results
Concluding Discussion

3. Background on Span-Protecting p-Cycles
[Grover and Stamatelakis, 1998]
“p-Cycles” = pre-connected, pre-configured, protection cycles
Operating Principle
BLSR-like loopback reaction under on-cycle span failure conditions plus additional protection against straddling span failures.
Add-ups over former span-protecting survivable networks:
Efficiency of mesh in the routing of working paths under no-failure conditions
Simple and fast switching as traditional SONET-rings in events of span failures
Improvements on spare capacity requirements for span-failure restorability purposes: a single p-cycle provides one restoration route to on-cycle spans and two protection segments to any straddling span.
Example of p-Cycle
Loopback reaction against on-cycle span failures
Break-in reaction against straddling span failures

4. Conventional p-Cycle Network Design (Context)
e.g. COST239 long-haul European network of 11 nodes, 26 spans (average nodal degree of 4.73) and 3531 candidate cycles.
Traffic matrix consists of 55 demand-pairs uniformly distributed on [1..20].
Resulting working capacities following a shortest distance based routing shown atop edges.
MIL
VIE
ZUR
PAR
LUX
BRU
LON
AMS
COP
BER
PRA 5 6 1 2 11 7 10 . 3 9 8 4
PAR MIL ZUR PRA VIE BER AMS LUX BRU LON COP
PAR
MIL
ZUR
PRA
VIE
BER
AMS
LUX
BRU
LON
COP 5 7 14 26 13 11 17 18 37 24 22 2 8 16 9 3

5. Conventional p-Cycle Network Design (ILP Model and Solution)
Objective:
Minimize spare capacity.
Optimal Solution:
55% of redundancy.
9 distinct structures: 4 are Hamiltonians.
16 unit-capacity p-cycles.
Constraints:
Number of channel copies of p-cycle p required.
2 copies
3 copies
1 copy
Full restorability under single span failures. 5

6. The COST239 Network under Single Node Failure Conditions
Node (instead of span) Failure View of the Problem
Number of paths affected in the event of node failure, e.g. Berlin has the most impact with 75 paths.
Number of paths transiting (and thus potentially restorable) the failed node, e.g. no more than 6 paths over 75 can be restored if a failure occurs in Berlin.
No transiting paths in Milan, Vienna, London and Copenhagen.
Total of 471 affected paths of which 119 may be considered for restoration.
MIL
VIE
ZUR
PAR
LUX
BRU
LON
AMS
COP
BER
PRA 5 7 14 26 13 11 17 18 37 24 22 2 8 16 9 3
Objective: Protection of Transiting Traffic in the Event of Single Node Failures.
(R1-node = Protected/Transiting)

7. Node Failure Restoration Options in p-Cycle “Span-Protected” Networks (1)
Inherent protection by a BLSR-like reaction to on-cycle node failures
[D. Stamatelakis and W. Grover, 2000], [D. Schupke, 2005].
Ordinary cycles may provide the same protection as BSLR-rings in the event of on-cycle node failures.
But benefits of p-cycles come from straddling span protection: What about “straddling” node failures?
Node-Encircling p-Cycles (NEPCs)
[D. Stamatelakis and W. Grover, 2000], [J. Doucette, P. Giese and W. Grover, 2005].
A p-cycle is considered to be an NEPC for a given node if it contains all neighbor-nodes of the former-node without crossing that (encircled) node itself.
Any flow transiting a node therefore intersects its NEPCs in a t least two of its neighbor-nodes and is then restorable by break-in reaction.
NEPC does not exploit “on-cycle” segment protection, Capacity-inefficient, relative scarcity of strict NEPCs, May require the usage of non-simple cycles, Complex ILP integrating cycle and p-cycle/NEPC enumeration.

8. Node Failure Restoration Options in p-Cycle “Span-Protected” Networks (2)
Path-segment Protecting p-Cycles
[G. Shen and W. Grover 2003, D. Onguetou and W.D. Grover 2008],
Really capacity-efficient,
Initially proposed for both node and span protection,
More complex failure detection and backup activation,
Basis of the present study (coming soon).
Failure-Independent Path-Protecting (FIPP) p-Cycles
[A. Kodian and W. Grover, 2005]
Operate like ordinary p-cycles for sets of end-to-end paths, which are mutually failure-disjoint between nodes of the FIPP.
Since path-oriented, FIPP p-cycles may provide an inherent protection to intermediate node failure as well, under the condition that the “disjointedness” is extended to nodes.

9. A Novel 2-Hop Flow Approach
Main Idea:
A given cycle may handle all traffic transiting through the common node of any 2-hop segment intersecting the cycle structure at its end-nodes.
Four Case Diagram:
Basic operating principle assuming the 2-hop segment as …
“virtual on-cycle span failure” for any intermediate on-cycle node failure,
“virtual straddling span failure” for any intermediate off-cycle node failure.
Main Features:
Uses simple cycles only, within the pre-processed set of candidates: no scarcity of NEPCs, and no complexity of cycle enumeration in the ILP.
Straightforward failure detections and backup activations; simple and fully pre-defined reaction to node failures. But still nearly as capacity-efficient as full flow-protecting p-cycles.

10. A Novel 2-Hop Flow Approach
Main Idea:
A given cycle may handle all traffic transiting through the common node of any 2-hop segment intersecting the cycle structure at its end-nodes.
Four Case Diagram:
Basic operating principle assuming the 2-hop segment as …
“virtual on-cycle span failure” for any intermediate on-cycle node failure,
“virtual straddling span failure” for any intermediate off-cycle node failure.
BLSR-like behavior, full on-cycle failed flow
Failed flow partially on the cycle
Failed flow: on-cycle nodes and straddling spans
Off-cycle node failure, straddling flow

11. Mathematical Programming Aspects
Re-use fully general path-segment protection ILPs previously developed; but with proper restriction to 2-hop segments.
Characterizing an existing design from the perspective of maximizing surviving paths under single node failure conditions,
Bi-criteria ILP biasing the selection of p-cycles (involved in the least cost solution) for maximum R1-node,
Maximize R1-node under a certain given budget,
Design for both 100% span and node failure restorability.
Equivalent ILPs implemented for the nepcs and fipp strategies.
Specific case of very large network instances:
The ILP solver is limited with the number of variables,
Our strategy when facing this is to reduce the set of candidate cycles,
Combination of GA-methods and ILP to achieve better pre-select sets.

12. The GA-ILP Meta-Heuristic: How it works?
No matter the size of the initial population
Typically converge in 15 iterations
Same fitness
Selection, Crossover, Mutation.
yes no
New Generation
Indexes of all possible cycles
Evaluation
Form union of all distinct cycles in final population
P1…P|P|/n
ILP
Size of union set is still O(size of individual)
All pairs
Best n/2
Crossover
Mutation*
P|P|/n+1...P2*|P|/n
ILP n
ILP
Pk…Pk+|P|/n
ILP
Final Solution *
P|P|-|P|/n...P|P|
ILP

13. A Novel GA-ILP Meta-Heuristic—why it seems promising/interesting?
Specific Aspect of this Contribution:
Novel and effective combination of GA-methods with ILP, which seems to have (at least qualitatively) many features to recommend it for any large p-cycle problem involving the selection of a relatively few optimal pre-cross-connected protection structures from an almost infinite space of all possible cycles.
Pre-selection approach, but with GA-like attributes having the capability of breeding collections or combinations of high merit candidates working well together; while reflecting exact goals of the original problem through ILP.
Greater conceptual simplicity and entirely repeatable meta-heuristic that simply re-uses existing p-cycle models but through GA-aspects (FIPP, node failure protection, optical length control, flow p-cycles, NEPCs, etc.)
Easily extendable to cases where the set of all possible cyclical structures is not even enumerable.
High quality solutions: as shown in the table, once the space of all possible cycles is fully enumerable, solutions are even equivalent to what would be obtained if instances were solvable with the entire set of candidates.

14. Validation of the GA-ILP Heuristic—Sample of Results
Design solutions always within 1% of optimality for cases where the space of all possible candidates is fully enumerable and solvable with the ILP—test cases for up to 85,000 candidates (with 19 nodes and 40 spans).
Solutions better than what where obtained using column generation for prior cases and for problems with fully enumerable candidate cycles but impractical to import into an ILP—test cases of up to 388,000 candidates (for 30 nodes and 59 spans).
more recently, high solution quality for a 200-node test case challenge.

15. Case Studies Five Test Case Networks: Sample Results:
Havana—17|N|, 26 |S|, 135 |P|, 58|D| on [1..5], shortest distance-weighted routing; vs. hop-count routing; vs. 136 |D| on [1..100].
Cost239—11|N|, 26|S|, 3531 |P|, 55|D| on [1..10], shortest distance-weighted routing.
Italy—13|N|, 24|S|, 557|P|, 78|D| on [1..10] , shortest distance-weighted routing.
Bellcore—15|N|, 28|S|, 976|P|, 104|D| on [1..20] , shortest distance-weighted routing.
Euro—32|N|, 42|S|, 699 |P|, 323|D| on [1..2] , shortest distance-weighted routing.
Sample Results:
Node Traffic Payload Characterization—affected vs. transiting flows
Reference Minimum Spare Capacity Designs
The 2-hop Flow Approach:
Maximization of surviving paths under minimum spare capacity requirements
Extra spare capacity requirements for 100% R1-node restorability
Comparison of R1-node and additional capacity for full R1-node with Other Approaches (i.e., nepc, fully general flows and fipp).

16. Maximum R1-node under Minimum Spare Capacity
Network
Affected
transiting
Redundancy(%)-min capa
2-hop
nepc
flow
fipp
Havana
271 77 85
fipp: +0.9
74 i.e. 96%
77 i.e. 100%
Hop-count
263 69 81
fipp: +17
53 i.e. 77%
58 i.e. 84%
larger |D|
29,317
14,591
123
fipp: -0.4
12,709 i.e. 87%
13,177 i.e. 90%
12,860 i.e. 88%
Cost239
471
119 62
fipp: -11
92 i.e. 77%
2 i.e. 2%
94 i.e. 79%
98 i.e. 82%
Italy
1357
521 89
fipp: -14
468 i.e. 90%
2 i.e. 0.4%
457 i.e. 88%
Bellcore
1326
396 79
fipp: -5
338 i.e. 85%
5 i.e. 1%
329 i.e. 83%
Euro
2483
1497 95
fipp: -9
1287 i.e. 86%
83 i.e. 6%
1350 i.e. 90%
1272 i.e. 85%
Fipp has its own designs that typically require less spare capacity than the equivalent conventional p-cycle network design.
2-hop: Very high levels of node failure protection are achievable under minimum spare capacity costs
The 2-hop strategy is nearly as capacity efficient as fully general flows
Nepc shows the worst performance because of scarcity of simple nepcs
Difficult to conclude on fipp because of different reference designs

17. Max R1-node under Min Costs for Havana Network

18. Extra over Min Spare Capacity for Full R1-node
Network
2-hop
nepc—max R1-node
flow
fipp
Havana (32 nepcs covering 4 nodes)
0.89%
R1-node 14%
+39%
0.3%
0.92%
Hop-count
39%
R1-node 25%
+77%
34%
47%
larger |D|
21%
R1-node 26%
+11% 9%
23%
Cost239 (1735 nepcs covering all nodes)
17%
Full R1-node
+138%
15%
-7%
Italy (359 nepcs covering 11 nodes)
18%
R1-node 69%
+73%
31%
Bellcore (847 nepcs covering 8 nodes)
11%
R1-node 34%
+67%
10%
-0.37%
Euro (458 nepcs covering 9 nodes)
9.5%
R1-node 28%
+65% 7%
-5%

19. Conclusion and Future Work
A simple new strategy and design method to recover from single node failures using only ordinary (span-protecting) p-cycles.
In test case networks we can protect 77% or more of node transiting traffic without penalty over minimum capacity-cost designs (for span restoration only).
Can achieve full node and span restorability for slightly more spare capacity than that is required in the conventional span-restorable network design.
BEST TRADEOFF for simple failure detection, backup activation and minimum-costs: 2-hop segment protection against both on-cycle and off-cycle node failures.
Ongoing Research Directions:
Multiple QoP Aspects,
Multi-layer Planning on a Controlled-Oversubscription Basis.

20. Questions, Comments?
A New Insight and Approach to Node Failure Protection with Ordinary p-Cycles
Wayne D. Grover and Diane Prisca Onguetou
TRLabs and University of Alberta
11th International Conference on Transparent Optical Networks / 5th Reliability Issues in Next Generation Networks (ICTON/RONEXT)
June 30th 2009, Island of Sao Miguel, Azores, Portugal