Controlling a cluster of computing resources: the model free control approach - Inria - Institut national de recherche en sciences et technologies du numérique Access content directly
Master Thesis Year : 2021

Controlling a cluster of computing resources: the model free control approach

Abstract

The recent feedback control approaches adopted in cluster and le server management of High Performance Computing (HPC) systems all utilizes the model-based control approach; these model-based control laws inherit the striking aws associated with this approach which include system modelling errors, complex parameter estimation etc., hence the need to take an alternative approach in this study. In this work, we propose to use the model free control approach with the following features: 1) a 'virtual model' representing the unknown system dynamics, 2) Elimination of the need of complex system parameter estimation and 3) assurance of less computational costs. This study rst explores and adopts model free control using the intelligent PID control law, which assures both robustness and quality reference tracking in respect to time varying system dynamics. Results from both simulations and real time experiments were presented in this work to explore the intuitive abilities of this approach. From the results, the Model free control approach guaranteed good reference tracking property and robustness to external disturbances which are key performance metric in HPC applications. This MFC approach via an Intelligent Proportional controller showed a 16:51% improvement in cluster usage without overloading the le-server compared to the 15:81% improvement by the classical Proportional-Integrator controller
Fichier principal
Vignette du fichier
Cluster Resource Management_MFC Approach.pdf (1.5 Mo) Télécharger le fichier
Origin : Files produced by the author(s)

Dates and versions

hal-03292373 , version 1 (21-07-2021)

Licence

Public Domain

Identifiers

  • HAL Id : hal-03292373 , version 1

Cite

David N Donkor. Controlling a cluster of computing resources: the model free control approach. Distributed, Parallel, and Cluster Computing [cs.DC]. 2021. ⟨hal-03292373⟩
81 View
97 Download

Share

Gmail Facebook Twitter LinkedIn More