Abstract
Traditional computing systems use simple, fast-to-compute heuristics to inform various decisions, such as allocating fractions of input/output bandwidth when switching among various tasks. Although fast and simple to execute, heuristics alone fail to account for the context of the computing system and the system’s behavior as a whole and cannot make a more-optimal decision that accounts for hardware components, energy resources, or the environment in which the computing device operates. A machine-learning-assisted or neural-network-based scheduler can make inferences or predictions based on the system information, conditions, and dynamics and make a more-optimal decision in allocating system resources during input and output operations. However, a machine-learning-assisted or neural-network-based scheduler can be slow to run, which may interfere with or disrupt normal computing system operations without an appreciable benefit. A hybrid, machine-learning-assisted scheduler can combine insights from a machine-learning model into traditional heuristic rules in such a way that the computing system can make more-accurate predictions with respect to input and output operation scheduling while still enjoying the fast response time of traditional heuristics.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Schott, Mark, "Machine-Learning-Assisted Scheduler for Filesystem Input and Output", Technical Disclosure Commons, (September 07, 2018)
https://www.tdcommons.org/dpubs_series/1492