Abstract
Imagine being able to dynamically create a computer with the exact resources a workload needs — with as much CPU or GPU power, or as much memory. This is the vision of composable systems. However, providers tend to implement in-house vendor-specific management APIs making integrations with platforms like Kubernetes increasingly more complex. Defining a common and usable method to manage composable resources is key. Ongoing initiatives like the Sunfish Management Framework — an open-source project from the OpenFabrics Alliance that provides a generic interface to query and manage fabric of composable resources — is an example of how the complexity of vendor-specific solutions can be abstracted. That said an additional integration step is required to bridge the gab between infrastructure management solutions like Sunfish and Kubernetes. In this talk we will describe the abstractions provided by Sunfish and we will present a Proof of Concept where we enable composable GPUs in Kubernetes.