Seminar Details
2026-10-06 (13:00) : Data Replication in geo-distributed systems
At shannon, Maxwell a.105
Duration: 30 minutes
Organized by Computer Science and Engineering
Speaker :
Cherif SI MOHAMMED (IMT Atlantique)
Abstract :
The presentation will be done in English.
The shift from Cloud Computing to Edge Computing has renewed the challenges of managing data in geo-distributed environments, particularly for large-scale data processing. Data replication, initially introduced to improve resilience and availability, is becoming increasingly costly due to growing data volumes and the resulting storage and bandwidth requirements. This thesis investigates dynamic replication strategies to reduce these costs while maintaining application performance. First, we study the state of the art in existing approaches for optimizing data replication and identify the main factors that support decision-making for adapting replication strategies. Second, we address the scheduling of Bag-of-Tasks applications that access shared data. We propose heuristic approaches that dynamically manage data replication and demonstrate, through comparison with exact methods, their ability to achieve near-optimal solutions while reducing data movement and resource consumption. Finally, as our heuristics make decisions for one job at a time based only on currently available resources, we investigate decision-making mechanisms to improve global system performance by jointly considering data and task migration across the geo-distributed infrastructure. Overall, this work contributes to more adaptive and resource-efficient data replication management strategies for Bag-of-Tasks applications in geo-distributed computing systems.