Ph.D. Defense: Kiranpreet Kaur – September 19, 2023

Kiranpreet Kaur, industrial Ph.D. student in collaboration with Orange Labs, will defend her Ph.D. thesis on September 19, 2023, at 9:30 a.m, in the Amphitheatre Georges Friedmann at Cnam, 2 rue Conté, 75003 Paris, France.

PhD Title: Dynamic migration and instantiation of microservices with end-to-end service chain performance guarantee

PhD Title (French): Migration et instanciation dynamique des microservices avec garantie de performance de bout en bout

Kiran worked on traffic management in network functions decomposed into microservices. She will work as R&D engineer at Dassault right after the defence.

Abstract:

The microservices and containerization plays an important role in the design of in the design of Virtualized Network Functions (VNFs), which are widely adopted by the telco-cloud industry. In practice, the cloud-native approach involves small and loosely coupled microservices that are deployed in containers and scaled (up and down) on distributed cloud servers/ data-centers. This is still a challenge to carefully orchestrate the allocation and rearrangement of (micro)services to avoid an in-balanced and a largely segmented solution space in a dynamic environment where services are arriving and leaving the network.Therefore, this PhD thesis aims to tackle the challenge of migrating and dynamic instantiation of containerized microservices in a distributed cloud architecture while maintaining the end-to-end service chain performance with the objective of low-latency exchange.To meet this objective, our first proposed contribution enables the latency-aware placement strategy for 5G/6G services over a substrate network. The strategy has been investigated from the perspective of their inter-dependency and the amount of traffic among microservices, which increased the service latency due to the delay associated with messages transiting through the transport network connectingdata centers. The proposed approach tends to minimize the global end-to-end latency, which is further solved using a hybrid heuristic algorithm and evaluated through the simulation experiments. Further, to allow the run-time placement of microservices and attain its optimality in a dynamic system, we rely on an approach that tends to trigger the dynamic migration and management of CNFs while considering the whole life cycle of containers on the basis of a driving use-case: an open-source 5G core network namely Magma. Here, we introduced three heuristic strategies to solve the formalized optimization model on migrating microservices across heterogeneous data center architecture. The evaluated results executed through the simulation shows better performance of the migration approachthat tends to minimize the global latency.Moving towards a more practical point-of-view, we performed an extensive study on various container-based migration techniques utilizing the popular orchestration tools (such as: Kubernetes, Docker Compose and Docker swarm etc.) to develop a Kubernetes based test-bed. A final work considered as a Proof-Of-Concept (PoC) has been developed to illustrate live migration of pods between remote Kubernetes clusters. As a use case, we consider the migration of a network function belonging to an open source 5G core network (namely, Magma).

Kiran’s publications

Access Thesis: https://theses.fr/2023HESAC014