Generalized multi-protocol label switching-based multi-layer networks (MLN) combining packet and optical switching lead to jointly leverage intrinsic per-layer benefits such as statistical multiplexing and huge transport capacity. By doing so, efficient network resource utilization is attained through MLN traffic engineering (TE) strategies, i.e. grooming. In this context, an optical link failure may cause the disruption of multiple groomed packet label switched paths (LSPs). Thereby, efficient recovery schemes such as restoration are required. In dynamic restoration, the centralized path computation element (PCE) sequentially computes backup paths for the set of failed packet LSPs using the TE database (TED). Since the TED is not updated until an LSP is actually set up, it is very likely that the PCE assigns the same network resources to different backup paths. This does increase resource contention and not fully exploits the potential grooming opportunities among the backup LSPs; consequently, the restorability metric performs poorly. To improve this, a designed PCE global concurrent optimization (GCO) architecture is implemented favoring grooming and lowering resource contention. The addressed problem, referred to as bulk path restoration in multi-layer optical networks (BAREMO), is formally modeled and stated using a mixed integer linear programming formulation. Then, a heuristic algorithm solving the BAREMO problem is devised. The experimental performance evaluation is conducted within the ADRENALINE testbed. Besides validating the PCE GCO architecture, its performance is compared with a sequential PCE for several traffic loads and failure rates. The results show that the PCE GCO improves remarkably restorability compared to the sequential PCE at the expenses, however, of increasing the restoration time.

Peer Reviewed