Multiple Failure Survivability with p-Cycles

Failure scenarios considered thus far, are “single failures”, and that means single fiber cuts or more generally, cuts of single edges of the network graph. In almost all survivability schemes, the objective is to replace the affected working paths in case of any single network failure. Networks that are fully restorable to single cuts are often called “100% restorable”. However, higher-order failure combinations can make such networks unable to fully recover. In other words, making a network fully restorable to single failures is not a guarantee that the availability of the service in the occurrence of higher order failures will be 100%. Several approaches have therefore been designed to improve the robustness of mesh transport networks against dual-failures. These approaches have either considered (pre-failure) strategies for addition of further protection capacity to achieve full or partial dual-failure survivability [55], [56] or have assumed reconfiguration of protection resources after the occurrence of the first failure to better withstand future failures [57], [58]. One alternative to cope with multiple failures in p-cycle based networks is p-cycle reconfiguration

that can be achieved by using static or dynamic (reconfigurable) p-cycles. Static p-cycle reconfiguration means that after a first failure the p-cycles remain as initially configured and the same set of p-cycles are used for recovering subsequent failures.

Conversely, in dynamic p-cycle reconfiguration the subsequent failures are recovered by finding new p-cycles in the remaining intact part of the network upon the first failure. Static reconfiguration is useful when dynamicp-cycle design is not possible or the reconfiguration after a first failure is not completed. These cases are considered in [59] and [60].

Usually the study of multiple failure survivable networks is simplified to consider-ing only dual failures, because occurrence of more than two failures at the same time is very unlikely [9]. The tradeoff between the number of deployed p-cycles and the survivability of dual fiber duct failures is investigated in [59]. In [60], it is assumed that dual failures are ordered events and the individual failures occur independently, such that the recovery of the first failure is completed before occurrence of the sec-ond failure. It should be noted that dual failure scenarios are only considered within one cycle, otherwise multiple failures can be protected by multiple separate p-cycles.

Results in [60] show that network designs with the minimal number of cycles and optimal capacity objectives are only able to restore around a half of the connections after the second failures. In [61], another mechanism called multi failure survivabil-ity (MFS) is introduced for recovering multiple failures one at a time. The results indicate that networks with higher average nodal degree are more likely to be sur-vived against multiple failures. Authors in [62] discuss the cases where the second failure occurs before recovering the first failure. Therefore, a fast readjustment of the p-cycles are required to temporarily protect the vulnerable working paths. The set of p-cycles can be redeployed either by a global optimization (where the whole network topology is readjusted) or by an incremental optimization (where only the vulnerable demands are re-protected by additional cycles).

In [63], the authors propose a method for dual-failure restoration by dynamically repairing p-cycles and compare it with incremental and complete dynamic recon-figurations. They studied the additional spare capacity required for dual failure restorability for each method and found that the efficiency of their dynamic repair method is in between complete and incremental reconfiguration schemes. It is clear that complete reconfiguration of p-cycles after the first failure is the most efficient method. Another article which discusses about p-cycle reconfiguration is [64] where the demands are divided into different service classes and dual failure survivability is provided to the highest priority demands called platinum traffic.

More recently, the authors of [65] have argued that, in addition to the above men-tioned approaches, reductions in the physical repair time of failures can also enhance service availability. They showed that an economic strategy exists for balancing the tradeoffs between capacity investment and Mean Time To Repair (MTTR) reduc-tion efforts to achieving high service availability in networks designed to be 100%

restorable against single failures. The authors of [9] studied the availability in span-restorable mesh networks. The availability analysis is based on the computational analysis of the restorability of a network to all possible dual-failure scenarios. In [66], the authors developed an analytical expression for the availability of paths in net-works using p-cycles as the protection mechanism. The model presented is based on the calculation of the unavailability caused by the effects of dual-failures and the authors have used the concept of “cutsets method” or “protection domain” to de-termine the service availability. An availability-aware service provisioning method in p-cycle based mesh networks is presented in [2]; therein, the service availability is analytically derived as a function of the span unavailability, using the concept of protection domain. The spare capacity is allocated, through a non-joint optimization model, to meet the availability requirement of the end-to-end traffic. More recently, this availability-aware network design method has been also applied for FIPP [67].