Thesis Code: 19009
Thesis Type: Master Thesis for Computer Engineering, Electronic Engineering
Research Area: Advanced Computing & Applications
- MS students in Electronic Engineering, Computer Engineering
- Experience with main programming languages (Python, C/C++, Go), algorithms
The pace at which Cloud based services are adopted is pushing Cloud service providers (CSPs) to adopt more heterogeneous resources in their data centers. Computing resource diversification is pushed also by the growing adoption of machine learning and deep learning algorithms (also HPC applications are going in this direction), which generally require specialized hardware to efficiently execute. Managing resources and tasks (i.e., deciding an allocation of the tasks under a certain set of constraints) at large scale requires innovative scheduling techniques. However, current schedulers still rely on simple strategies to assign tasks to available resources.
The objective of the work is to study innovative approaches for managing resources in Cloud-HPC environments. To this end, machine learning and evolutionary based techniques will be considered for improving the quality of task scheduling when heterogeneous resources are available (e.g., GPUs, FPGAs, dedicated ASICs). Studied techniques will look at improving scheduling under different constraints (energy saving, reducing task makespan, etc.) and combination of them. Targeted technologies include Linux containers and related orchestrators such as Kubernetes.