A Kubernetes-based Resource Augmentation Framework for Edge Devices
K-RAF assists edge devices to overcome their limited capabilities by provisioning virtualized computation and storage resources in a Kubernetes environment.
K-RAF consists of three modules (Execution Migration Module, Monitoring Module, and Storage Module):
-
Execution Migration Module:
- Utilizes migration features of K-RAF.
- Receives IoT device's app execution environment info from the database.
- Performs migration on the edge server by creating application pods on a Kubernetes cluster.
- User input for GPU usage guides GPU allocation during migration.
- Kubernetes' Custom Resource Definition partitions GPU VRAM for GPU-reliant applications.
-
Monitoring Module:
- Consists of Prometheus, Grafana, and Node-exporter.
- Node-exporter deployed as a pod on each node of the edge server to collect data.
- Prometheus stores collected data in a database.
- Grafana visualizes data from Prometheus via a dashboard.
- Users can query Prometheus to retrieve specific data.
-
Storage Module:
- Comprises K-RAF's volume-related functions and rook-ceph.
- rook-ceph builds a storage pool using HDDs and SSDs in the edge server cluster.
- SSDs classified as cache and HDDs as backing storage for cache-tiering.
- Kubernetes creates a PV using storage from the pool.
- K-RAF uses WebDAV Server pods to mount PV to IoT devices for virtual storage.
Kubernetes-based Resource Augmentation Framework
- [Kubernetes v1.22.9]
- [Docker v20.10.16]
- [rook-ceph v1.9.10]
- [davfs2 v1.6.1]
Edge Server Cluster
- [Intel i7-11700]
- [RAM 64 GB]
- [SSD 500 GB]
- [HDD 1 TB]
- [Ubuntu 19.04]
Edge Device
- [Raspberry Pi 4]
- Installed davfs2 as WebDAV client