PhD "Supervised Learning of Availability for Digital Twin of Infrastructures" M/F

Twoja rola

Your role is to conduct a thesis on: « Supervised Learning of Availability for Digital Twin of Virtualization Infrastructures. »

Global context and problematic of the subject
Hosting interruption-sensitive applications on distributed data centers, also known as "edge computing," requires the establishment of multi-site protection schemes and a deep understanding of the associated failure risks [1], [2]. Furthermore, the recent development of Digital Twins [3] for the automated operation of infrastructure networks necessitates the creation of network models and the implementation of automated information collection to synchronize the state of the digital twin with that of the infrastructure elements. The dynamic orchestration of virtualization containers (Kubernetes) and the associated monitoring architectures (Prometheus) open new perspectives for information collection and automatic adaptation in the face of potential infrastructure degradation [4], [5], [6].

Scientific Objective
The objective of this thesis is to model the failure statistics of infrastructure elements and to develop supervised learning of its parameters, with the aim of compressing measurements taken by infrastructure probes for a Digital Twin.

Challenges to Overcome
1)     Avoid the loss of short failure events during historical data collection through subsampling.

2)     Aggregate rare failure event statistics from a set of infrastructures in production.

3)     Build an availability model that generates synthetic parameters (availability, state probability).

4)     Prototype an exporter to be integrated into a Prometheus measurement chain.

References
[1] I. Narayanan, Right-sizing Geo-distributed Data Centers for Availability and Latency, 2017

[2] K. Sayad, Interdependency-Aware Resource Allocation for High Availability of 5G-enabled Critical Infrastructures Services, 2022

[3] A. Thelen, A Comprehensive Review of Digital Twin - Part 1: Modeling and Twinning Enabling Technologies, 2022

[4] D. Tazzioli, Stateful Service Migration Support for Kubernetes-based Orchestration in Industry 4.0, 2024

[5] T. Trung Le, Hidden Markov Models for diagnostics and prognostics of systems under multiple deterioration modes, 2014

[6] A. Samir, Self-Adaptive Healing for Containerized Cluster Architectures with Hidden Markov Models, 2019