Dynamic System Characterization of Enterprise Servers via Nonparametric Identification
E. Schuster and K. C. Gross
American Control Conference
Portland, Oregon, USA, June 2005
Abstract
|
|
Dynamic characterization and fault detection are carried out in
enterprise servers using nonparametric identification techniques based
on sinusoidal excitation. The introduction of subtle sinusoidal
perturbations in computer load variables or physical variables allows
us to obtain a dynamic input-output characterization in the frequency
domain. The input-output relationship is described in terms of
coupling coefficients between a wide variety of physical and
performance variables at different selected frequencies. This
innovative approach in the field of computer science has been
demonstrated in empirical studies to provide valuable dynamic system
characterization information that can be indispensable to datacenter
operations personnel for the functions of performance management,
capacity planning, quality-of-service (QoS) assurance, dynamic
resource provisioning, and root cause analyses.