Autoscaling is critical for cloud service providers. Precision and reactivity are criteria sought in order to avoid extra financial costs. Indeed, the over-allocation of resources generates a cost of purchasing and using more machines. On the other hand, the under-allocation leads to a degradation of the QoS which results in violations of the SLAs.
In the containers domain, the most popular solution is horizontal scaling of Kubernetes (HPA). In this work, we propose two hybrid scaling algorithms benefiting from the advantages of both horizontal and vertical scaling. In addition, our second algorithm uses a profiler able to estimate the CPU requirement to respond to a certain number of received requests. Our algorithms have been tested in different situations, and their performances are compared to those of Kubernetes HPA. The results obtained demonstrate a significant improvement in QoS, resources usage and allocation, as well as in energy consumption.
| Date | 6 May 2021 |
|---|
| Original language | French |
|---|
| Awarding Institution | - École de technologie supérieure
|
|---|
| Supervisor | Abdelouahed Gherbi (Supervisor) & Nadjia Kara (Co-supervisor) |
|---|
Ben Cheikh Larbi, A. (Author),
Gherbi (Supervisor) &
Kara (Co-supervisor),
6 May 2021Student thesis: Master's thesis › Master in Engineering: Information Technology Engineering