This experimental scenario has now been published!
This scenario has been presented on the IEEE Cluster Computing 2020, if you need to cite us, you can use:
Jonatan Enes, Guillaume Fieni, Roberto R. Expósito, Romain Rouvoy, Juan Touriño. Power Budgeting of Big Data Applications in Container-based Clusters. 2020 IEEE International Conference on Cluster Computing (CLUSTER), Kobe, Japan, 2020, pages 281-287. Online | Preprint | Video
The Energy capping scenario and functionality is created by using both the BDWatchdog monitoring capabilities as well as the Serverless Containers framework. This combination makes it possible to rescale the CPU in real time and thus adjust the energy that a container or a set of containers consume at any moment. In addition, in order to have energy metrics a key tool, PowerAPI, is also used.
With this energy management feature some interesting features can be exploited:
- Energy as another accountable resource, as now energy can be seen as a resource and thus it can be shared and accounted between the users/containers.
- Energy limit enforcement by setting a limit to a container/group of containers/user and having it respected.
There is a public demo where several plots can be seen for experiments where the energy capping features were used.
You can see them by clicking on the Demo section or with this link.