Adrian Khelili

– Thesis: Intelligent File Placement in Hierarchical Storage Through File Lifecycle Exploitation for HPC – As part of the Data Management team, I focused on resolving I/O bottlenecks within burst buffer systems used in high-performance computing. My role involved applying Machine Learning techniques, particularly time series forecasting, to enhance data storage efficiency and system performance. – I developed and implemented methods for optimizing data placement across different storage tiers, including RAM, SSDs, and HDDs. This work involved creating strategies for data prefetching and eviction to ensure rapid access to frequently used data while maintaining efficient storage utilization.

– As part of the Data Management team, I developed FiLiP, an advanced tool for optimizing file placement policies in hierarchical storage systems used in high-performance computing. My work focused on visualizing file lifecycles and movements across storage tiers, including RAM, NVMe, and HDDs, to identify and resolve I/O bottlenecks. Key achievements: – Implemented methods for analyzing and improving file placement efficiency, leveraging time series forecasting and data visualization techniques. – Developed a web-based interface with Vue.js and D3.js, and a REST API using FastAPI, to provide real-time insights and metrics on storage utilization and policy effectiveness.