Deploy

dspk serve turns a GPU machine into a server. It holds the recordings, answers views from their min/max pyramids, and runs sorts on its GPU. Clients connect over QUIC with a certificate the server issued them: the command line, the app and Python all work the same way.

Follow these in order the first time:

1. Build the image: one Containerfile, a target per kind of GPU.
2. Run the server: your folder of recordings at /data.
3. Enroll clients: who may connect, and who may run jobs.
4. Work from a client: list files, read, sort, fetch.

Then, as needed:

• Pass the GPU in, including SELinux on Fedora and RHEL.
• Run it as a service that starts at boot.
• Security: what protects the server, and what doesn’t yet.
• Smoke test: check an image end to end.