Abstract
Modern cluster management systems have effectively evolved to deal with the increasing and diverse cloud computing demands. However, several challenges including low resource utilization, high power consumption are still present that can be solved with a precise real-time usage prediction. This prediction problem is complicated since the cloud workloads vary dynamically and there are nonlinear relationships between the usage, duration and jobs characteristics. Therefore, nonlinear feature extraction methods including logarithm, encoder and several feature extraction methods were used in the past studies. Our study utilized several regression models and deep learning models including GRU, LSTM in univariate and multivariate settings to explore and extract highly-dimensional and highly-nonlinear relationship. Our experiments used Google Cluster Trace data v3 to perform prediction on duration, CPU and memory utilization.
Original language | English |
---|---|
Title of host publication | Proceedings of the 2023 International Conference on Advances in Computing Research (ACR '23) |
Editors | Kevin Daimi, Abeer Al Sadoon |
Place of Publication | Switzerland |
Publisher | Springer |
Pages | 443-452 |
Number of pages | 10 |
ISBN (Electronic) | 9783031334437 |
ISBN (Print) | 9783031337420 |
DOIs | |
Publication status | Published - 2023 |