Red Hat approach
With Red Hat® platforms, NASA and its contractors can mix and match edge hardware, based on space, power, cooling, and connectivity constraints on space stations and satellites. Options include remote worker nodes and three-node clusters.
To start, build ML models on Red Hat OpenShift®, which packages the code with everything it needs to run, including the operating system, tools, and libraries. One advantage of containerized applications in space is their size—typically a few megabytes compared to gigabytes for an equivalent virtual machine. Another advantage is portability. Researchers can develop an ML model once and then deploy it on any hardware—at the edge, in the cloud, or spread across both (the hybrid model). During peer review, other scientists can reproduce the results on any platform, regardless of operating system, libraries, or system configuration.
To ingest data from diverse sensors in space, use Red Hat AMQ streams. Based on the sensor data, Red Hat OpenShift Serverless scales the application up or down. When no data is coming in, Red Hat Ansible® Automation Platform turns off the model to conserve power and cooling. When sensor data begins streaming again, Ansible Automation Platform dynamically spins up the right number of instances of the model to meet performance targets.
Red Hat provides the 24x7 support needed to support NASA’s mission. We extensively test software from the open source community before distributing it. And our engineers work continually to improve features, reliability, and security to make sure the software infrastructure remains stable in high-stakes environments like space.