Prédiction des catastrophes naturelles et survivabilité contre les pannes liés aux catastrophes dans les réseaux de centre de données optiques

Enhancing survivability for elastic optical inter-DataCenter networks

• Auteur / Autrice : Yuanhao Liu
• Direction : Fen Zhou, Juan-Manuel Torres-Moreno
• Type : Thèse de doctorat
• Discipline(s) : Informatique
• Date : Soutenance le 14/11/2022
• Etablissement(s) : Avignon
• Ecole(s) doctorale(s) : École doctorale Sciences et agrosciences (Avignon)
• Partenaire(s) de recherche : Laboratoire : Laboratoire informatique d'Avignon
• Jury : Président / Présidente : Abderrahim Benslimane
• Jury : Examinateurs / Examinatrices : Ahmed Meddahi, Tao Shang
• Jury : Rapporteur / Rapporteuse : Catherine Lepers, Abderrezak Rachedi

Mots clés

Mots clés contrôlés

Optimisation globale

Protection

Réseaux de diffraction

Résilience

Prévision

Catastrophes naturelles

Mots clés libres

Prédiction des catastrophes naturelles

Survivabilité

Pannes liés aux catastrophes

Réseaux de centre de données optiques

Network survivability

Elastic Optical Inter-DataCenter Networks

Directed p-cycle protection

Integer linear program (ILP)

Dedicated path protection

Résumé

Non communiqué

With the widespread popularity of applications such as cloud computing, ultra-high-definition video-on-demand, streaming, etc., the exponential growth of the generated information presents a challenge for the current optical networks, especially for Inter-DataCenter (DC) networks, which carry massive amounts of DC-to-client information. The traditional wavelength division multiplexing (WDM) optical network is not adapted to future networks due to its ''one-size-fits-all'' data transmission mode. Meanwhile, elastic optical networks (EONs) have shown the flexibility and scalability to solve the problem of huge traffic provisioning. It is widely regarded as the next-generation optical network architecture and inter-DC network, which is the so-called Elastic Optical Inter-DataCenter Network (EO-DCN). However, the multitudes of Internet services are facing critical threats from network failures, e.g. fiber cut, and natural disruption. These network failures will cause incalculable financial losses. Thus, network resilience is one of the most significant factors for a majority of DC-based services. In this dissertation, we develop two classic protection schemes in EO-DCNs, i.e. directed pre-configured cycles (p-cycle) protection and dedicated path protection, against single link failure and disaster failure, respectively, to further enhance the survivability of EO-DCNs. We first propose to investigate how to schedule the directed p-cycle protection for lower power consumption based on a compact modulation format selection in EO-DCNs. Instead of a brief upper bound on the modulation format assignment in the conventional directed p-cycle design, the proposed modulation format adaptation is designed as just enough for the directed p-cycle, which is determined by the distance of each on-cycle protection path. The problem involves the directed p-cycle generation, modulation format adaptation, power consumption minimization, and spectrum allocation. To this end, three different integer linear programs (ILPs) are formulated to provide joint optimization of power consumption and spectrum usage for all the directed p-cycles generated in EO-DCNs. The directed p-cycles are constructed via flow conservation in a way without candidate cycle enumeration, to reduce computational complexity. Then, the modulation format adaptation is achieved by relying on different flows in ILPs. To solve the problem with large-scale instances, the column generation (CG) approach is proposed to construct a promising solution. To guarantee the performance with respect to the optimal solution, the epsilon-accuracy is then introduced to evaluate the optimality of the CG approach. Extensive simulations are conducted to compare the proposed ILP and the CG approach with the conventional ones. Numerical results demonstrate that the proposed ILP models show better performance on power consumption. The proposed CG approach is also proven with high computation efficiency and guaranteed performance with large-scale traffics. We then investigate disaster protection for EO-DCNs. We focus on the disaster-resilient service provisioning problem leveraging cooperative storage system (CSS). Instead of mirrored content backup on a single DC, our proposed CSS partitions a required content into no less than three fragments if possible, each of which is then stored on a DC located in different disaster zones (DZs), to ensure they are DZ-disjoint. Accordingly, multi-path routing with the adaptive number of working paths to distinct DCs is employed to serve each request, while a protection path is computed to protect against a disaster failure. The studied problem involves DC assignment, content partition, and placement, working/protection paths computation, as well as spectrum allocation. Our main objective is to jointly minimize the spectrum usage and maximal occupied frequency slot index (MOFI) subject to disaster resilience.