Evaluating Cloud Auto-Scaler Resource Allocation Planning Under Multiple High-Performance Computing Scenarios

Leochico, Kester
Journal Title
Journal ISSN
Volume Title

Cloud computing enables users to elastically acquire as many computing resources as they need on-demand while minimizing resource allocation to what is needed on the provider side, allowing users to acquire only the resources that they need while reducing the costs of acquiring said resources over traditional datacenters. The key to enabling these improvements are cloud auto-scalers, which are subsystems that are responsible for planning out how many resources to provision in response to current/future demand. The current state of the art in terms of comparing different auto-scaling algorithms is immature, however, due to the lack of consistent, standard evaluation methodologies that analyze cloud auto-scalers under multiple scenarios, making it difficult to compare the results between proposed auto-scalers. In an effort to address some of these issues, this work analyzes the effects of changing the workload mix and lowering the average job runtime (as represented by the service rate) on the performance of three cloud auto-scalers from the literature in an effort to better model the behavior of cloud auto-scalers under high-performance computing scenarios involving long-running jobs rather than short-lived jobs as in previous studies. The simulation and analysis was carried out using the Performance Evaluation framework for Auto-Scaling (PEAS), a cloud simulator framework for modelling the performance of cloud auto-scaling algorithms using scenario theory.

This item is available only to currently enrolled UTSA students, faculty or staff.
Auto-scaling, Cloud computing
Electrical and Computer Engineering