UDC 004.942
COMPARATIVE STUDY OF LINEAR REGRESSION METHODS TO BUILD ADAPTIVE MANAGEMENT SYSTEM IN HIGH-PERFORMANCE VIRTUALIZATION ENVIRONMENTS
A. O. Ferubko, post-graduate student, BSTU, Bryansk, Russia;
orcid.org/0009-0006-5625-137X, e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
O. D. Kazakov, PhD, Associate Professor, Head of the Department, BSTU, Bryansk, Russia;
orcid.org/0000-0001-9665-8138, e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
The problem of predicting RAM consumption in a virtualized environment based on KVM/QEMU for the purposes of adaptive resource management, RAM in particular, in a homogeneous system is considered. The aim is to find the main performance indicators of linear regression models for predicting RAM consumption of both guest and host operating systems (OS). The forecast is based on a training sample, which is formed from data on RAM consumption at previous points when the system made a decision on RAM redistribution. The decision-making model is based not on the consumption forecasts themselves, but on the volumes of free (unused) RAM, the value of which is obtained by subtracting from the entire RAM subsystem the RAM that is predicted to be involved in subsystem processes. And based on the forecast of unused RAM, an informed de cision about the need and scale of RAM redistribution between subsystems is made. The decision depends not only on forecasts, but also on the size of the window that protects the system from too frequent redistri butions in conditions of relatively small difference, as well as on the OS themselves and their restrictions on minimum amount of RAM.
Key words: RAM, KVM, QEMU, hypervisor, virtualization, hypervisor of the first type, linear regression, regularization, Tikhonov L2-regularization, L1-regularization, weighted linear regression, time series, oper ating systems, homogeneous systems, machine learning.