Andreas Kirstädter | Searcharxiv

arXiv2023

Tactile Network Resource Allocation enabled by Quantum Annealing based on ILP Modeling

Agile networks with fast adaptation and reconfiguration capabilities are required for on-demand provisioning of various network services. We propose a new methodical framework for short-time network optimization based on quantum computing (QC) and integer linear program (ILP) models, which has the potential of realizing a real-time network automation. We define methods to map a nearly real-world ILP model for resource provisioning to a quadratic unconstrained binary optimization (QUBO) problem, which is solvable on quantum annealer (QA). We concentrate on the three-node network to evaluate our approach and its obtainable quality of solution using the state-of-the-art quantum annealer D-Wave Advantage 5.2/5.3. By studying the annealing process, we find annealing configuration parameters that obtain feasible solutions close to the reference solution generated by the classical ILP-solver CPLEX. Further, we studied the scaling of the network problem and provide estimations on quantum annealer's hardware requirements to enable a proper QUBO problem embedding of larger networks. We achieved the QUBO embedding of networks with up to 6 nodes on the D-Wave Advantage. According to our estimates a real-sized network with 12 to 16 nodes require a QA hardware with at least 50000 qubits or more.

quant-ph↗

arXiv2023

Grooming Connectivity Intents in IP-Optical Networks Using Directed Acyclic Graphs

During the last few years, there have been concentrated efforts toward intent-driven networking. While relying upon Software-Defined Networking (SDN), Intent-Based Networking (IBN) pushes the frontiers of efficient networking by decoupling the intentions of a network operator (i.e., what is desired to be done) from the implementation (i.e., how is it achieved). The advantages of such a paradigm have long been argued and include, but are not limited to, the reduction of human errors, reduced expertise requirements among operator personnel, and faster business plan adaptation. In previous work, we have shown how incorporating IBN in multi-domain networks can have a significantly positive impact as it can enable decentralized operation, accountability, and confidentiality. The pillar of our previous contribution is the compilation of intents using system-generated intent trees. In this work, we extend the architecture to enable grooming among the user intents. Therefore, separate intents can now end up using the same network resources. While this makes the intent system reasonably more complex, it indisputably improves resource allocation. To represent the intent relationships of the newly enhanced architecture, we use Directed Acyclic Graphs (DAGs). Furthermore, we appropriately adapt an advanced established technique from the literature to solve the Routing, Modulation, and Spectrum Assignment (RMSA) problem for the intent compilation. We demonstrate a realistic scenario in which we evaluate our architecture and the intent compilation strategy. Our current approach successfully consolidates the advantages of having an intent-driven architecture and, at the same time, flexibly choosing among advanced resource allocation techniques.

cs.NI↗