Scheduling Algorithms for Multiprogramming in a Hard-Real-Time Environment
Journal of the ACM (JACM)
ESP: a system utilization benchmark
Proceedings of the 2000 ACM/IEEE conference on Supercomputing
Introduction to Reinforcement Learning
Introduction to Reinforcement Learning
Neuro-Dynamic Programming
Scheduling dependent real-time activities
Scheduling dependent real-time activities
HPC Productivity: An Overarching View
International Journal of High Performance Computing Applications
High Performance Computing Productivity Model Synthesis
International Journal of High Performance Computing Applications
Utility accrual real-time scheduling: models and algorithms
Utility accrual real-time scheduling: models and algorithms
Reinforcement learning: a survey
Journal of Artificial Intelligence Research
Planning and acting in partially observable stochastic domains
Artificial Intelligence
A reinforcement learning framework for online data migration in hierarchical storage systems
The Journal of Supercomputing
Hi-index | 0.00 |
This paper presents a general methodology for scheduling jobs in soft real-time systems, where the utility of completing each job decreases over time. This scheduling problem is known to be NP-hard, requiring a heuristic solution to operate in real-time. We present a utility-based framework for making repeated scheduling decisions based on dynamically observed information about unscheduled jobs and system's resources. This framework generalizes the standard scheduling problem to a resource-constrained environment, where resource allocation (RA) decisions (how many CPUs to allocate to each job) have to be made concurrently with the scheduling decisions (when to execute each job). We then use the discrete-time Optimal Control theory to formulate the optimization problem of finding the scheduling/RA policy that maximizes the average utility per time step obtained from completed jobs. We propose a Reinforcement Learning (RL) architecture for solving the NP-hard Optimal Control problem in real-time, and our experimental results demonstrate the feasibility and benefits of the proposed approach.