Scheduling for Accelerated Devices
The GPU Scheduler tool determines the best scheduling and GPU allocation for Deep Learning training jobs, reducing energy and execution costs (in both private or public clouds) while meeting deadline constraints. The tool only requires the list of submitted jobs (provided as Docker containers) with their characteristics (expected execution times, priorities, and deadlines) and a description of the available resources in the system.
Automatically performing resource selection and scheduling, allows resource usage optimisation, reducing the amount of idle GPUs with a negative impact on the overall costs (operational in case of public cloud, energy in case of public clouds). The optimal assignment reduces the probability of violating jobs deadlines, thus minimising the corresponding penalties.
A GPU scheduler to optimise the execution of Deep Learning training jobs on GPU-accelerated clusters, minimising the operational costs and the due dates violations