Application Interconnect demo

[scottmbaker]

I've recorded a demo of EMF's Application Interconnect feature and it has been added to the Intel youtube channel.

The Interconnect is a feature that we have not talked about much, yet there has been some interest in the capability from partners, so I created this video that records the steps necessary to deploy applications using the Interconnect feature. The purpose of the Interconnect is to provide a secure L7 application mesh between edge clusters.

Our original vision was to interconnect AI training and inference, so that inference deployed to the distributed edge (for example in retail or manufacturing situations) could connect to a centralized training application residing at a different edge. For example a thousand retail edges reaching out to a training cluster to obtain model updates. Hence it was designed to naturally facilitate hub-and-spoke patterns, though it's not necessarily limited to that only that pattern.

The implementation leverages a component called Skupper. Skupper provides the routing functionality at the edge as well as the means to create links between multiple edges and the ability to expose services from one edge to another. If a service is exposed one edge1, and edge1 is linked to edge2, then that exposed service will appear as a local service on edge2. All the routing will be handled by the interconnect. Skupper secures the connections with mTLS. This means that applications may be written agnostic to complexities of edge-to-edge networking.

What EMF's Interconnect adds on top of the basic Skupper functionality is automatic configuration and management of the links and the exposed services. EMF exposes a network abstraction. If you create a network and you attach two or more EMF deployments to that network, then EMF will automatically reach out to Skupper to link those applications together. This is implemented by a Kubernetes operator, so EMF is constantly watching the set of deployed applications and managing the Skupper links as necessary. Because EMF tracks the applications that participate in the network as well as which services are exposed, EMF can automatically manage topologies such as hub-and-spoke, only creating links between the edge clusters that need to talk to one another.

The demo is here and will take you through deploying a load-balancer, deploying Skupper, creating a network, and deploying a pair of sample applications in a client-server configuration to use that network.

If you find this feature useful, please let us know, and let us know what innovative use cases you may have for connecting multiple edge clusters.