Cloud computing systems are known for their elastic property which allows the dynamic addition and removal of resources based on the on-demand service model. However, the unsteady workloads and the variety of the applications running in the cloud entail the problems of resources over-provisioning and under-provisioning, which cause resources wastage and user experience degradation. In order to address this problem, workload modeling can be used to design proactive decision-making approaches to optimize the resource provisioning strategies and anticipate any performance problem.
Workload models are typically built based on user and application behavior in a system, limiting them to specific domains. Undoubtedly, such practice forms a dilemma in a Cloud environment where a wide range of heterogeneous applications are running, and many users have access to these resources. The workload model in such infrastructure must adapt to the evolution in the system configuration parameters like job load fluctuation, horizontal scaling, vertical scaling, or migration of virtualized resources. Moreover, the collected workload data often hold precious information about recurring patterns in the evaluated system’s resource behavior. Classification of such periodic patterns by amplitude, length and shape can be an invaluable asset to researchers aiming to improve workload models.
The aim of this work is (1) to generate generic workload models which are independent of user behavior and the applications running in the system, (2) that are able to fit any workload domain and type, (3) that are able to model sharp workload variations that are most likely to appear in cloud environments, and (4) with high degree of fidelity with respect to the observed data in a short period of execution time. As a subset, a workload periodicity detection mechanism is also proposed, enabling detection of cyclic workload patterns and workload behavior anomalies. To achieve this aim, this work is therefore divided into two complementary working areas. The first working area centers around two approaches for workload estimation, while the second working area centers around workload periodicity detection.
| Date | 8 Jun 2018 |
|---|
| Original language | French |
|---|
| Awarding Institution | - École de technologie supérieure
|
|---|
| Supervisor | Nadjia Kara (Supervisor) |
|---|
St-Onge, C. (Author),
Kara (Supervisor),
8 Jun 2018Student thesis: Master's thesis › Master in Engineering: Information Technology Engineering