Stochastic Preplanned Restoration Algorithms

The ability of overcoming service interruptions, i.e., resilience, has always been an irremissible requirement for communication networks. Resilient schemes can either reserve in advance network spare resources (protection schemes) or found them upon failure occurrence (restoration schemes). While fixed protection schemes have been extensively used thus far in the telephone industry, the envisioned dynamic Optical Layer (OL) and the lack of flexibility proper of protection schemes are driving network designer to resort to restoration schemes. By dynamically looking for backup paths of spare wavelengths upon failure occurrence, restoration schemes have the potential to yield efficient and flexible resource reservation. However restoration schemes, generally, present long failure recovery time (i.e., the time required to restore the disrupted connections). This is mainly due to the heavy signaling that originates upon failure occurrence. In high capacity WDM networks, the presence of many connections concurrently seeking restoration exacerbates the above problem as, in existing restoration schemes, coordination among restoration attempts may further slow down the process completion. To exploit the flexibility of restoration schemes and decrease their failure recovery time and resource contention, the authors propose a class of fast and efficient path restoration schemes called Stochastic Preplanned Restoration (SPR) schemes. In the SPR schemes each active connection is associated with an agent resident at the connection master node. Communication between agents is not allowed. This permits to reduce the coordination required during the failure recovery process. Distinct restoration paths are precomputed at the connection master node at the time the connection is set up. Upon failure of the connection working path, one of the preplanned paths is randomly selected as restoration path depending on specific probabilities calculated by the connection agent. The proposed schemes require limited signaling upon failure occurrence as coordination among the agents involved in the restoration process is not required. Therefore, the schemes are fast and scalable in terms of number of network nodes, link or fiber capacity, and number of connections. Yet, presented results show that the SPR schemes may considerably decrease the blocking probability of the restoration attempts when compared to other distributed restoration schemes, such as alternate routing, and perform closely to centralized schemes based on ILP solutions.