| |
Click here for full text:
Characterizing Application Workloads on CPU Utilization for Utility Computing
Abrahao, Bruno; Zhang, Alex
HPL-2004-157
Keyword(s): capacity planning; principal component analysis; workload model; trace characterization and generation
Abstract: We analyze CPU utilization traces of multiple applications running on a shared set of processors in a utility computing environment and apply PCA (Principal Component Analysis) technique to characterize each application's workload. We show that, in our dataset, the 12 applications under examination can be characterized by just three features, namely, periodic, noisy, and spiky. We then use these principal components for classifying applications, detrending the CPU usage behavior, and generating synthetic traces with amplification or suppression of the desired features. The workload characteristics that we derive using the PCA approach help application owners to better understand the behaviors of their applications, and also enable the system operator to better plan for capacity usage.
8 Pages
Back to Index
|
Printable version
