Instead of relying on a centralized control plane for edge infrastructure, we propose to use independent OpenStack clouds that collaborate if needed. An approach that enables the segregation of the infrastructure into distinct areas resolving network partitioning and scalability challenges in most cases.
The main idea consists in extending the OpenStack CLI to define the collaboration process. A dedicated option, called --scope, specifies which services (e.g., compute, image) of which OpenStack clouds (e.g., Cloud1, Cloud2) an OpenStack is made of to perform the CLI request. For instance, the provisioning of a VM in Cloud1 with an image in Cloud2 looks like as follows:
openstack server create vm --flavor tiny --image cirros \
--scope '{"compute": "Cloud1", "image": "Cloud2"}'
First, we will explain the limitations of a centralized control plane. Then, we will present the --scope proposal and its proof-of-concept: a simple proxy in front of OpenStack clouds.
Proposing an edge resource management system is a hot topic. Since 2016, works have been investigating the use of a single OpenStack cloud deployed WANWide (i.e., a centralized control plane and remote compute nodes). Such an approach is relevant for specific use cases (e.g., virtual offices). But, it does not address network partitioning issues that occur in large-scale scenarios.
Leveraging the initial prototype developed during the Berlin Hackathon, this talk presents an innovative approach that enables on-demand collaboration between independent OpenStack deployments. It keeps each edge site independent and makes them collaborating only if needed, which achieved a scalable and robust way to operate and use edge infrastructures.
Attendees may expect to learn:
- Limitations of a centralized control plan
- Challenges/Opportunities of on-demand collaborations
- The solutions used to implement our proof-of-concept and the current limitations