Containers and Virtualization in the Cloud

The container technology can provide lightweight virtualization with the aid of operating systems and is being rapidly adopted to emerging cloud systems.

Resource Virtualization for Containers

Linux control group (cgroup) provides the framework to control resources assigned to each task group. In the container-based cloud, a container is mapped into a task group in Linux. Accordingly, the resource attributes of containers are ultimately mapped into those of cgroup. To provide QoS for best-effort jobs in the container-based cloud, we propose the weighted CPU resource management for low-priority task groups while preventing resource interference with high-priority jobs [CCPE2023]. We are also studying on fine-grained virtualization of high-end devices, such as GPU and InfiniBand, for containers.

Resource Provisioning and Autoscaling

Resource autoscaling can be classified into horizontal autoscaling and vertical autoscaling. Horizontal autoscaling automatically scales the number of containers, while vertical autoscaling scales the amount of resources reserved by a container. We are studying on vertical autoscaling of GPU and CPU with consideration of application level QoS [BigData2020, ComNet2023]. Our preliminary performance measurement results show that our approach can provide optimal performance to containerized CNN-based machine learning applications, such as ResNet-50 and Yolo-v4.