Could you please provide some details on how shield actually works? I am planning on working on a similar project for personal / educational purpose. It involves remote/headless OpenGL rendering to concurrent context(s)/buffers. However there is not much specific information regarding such topics. OSMesa does not seem to be documented at all - it is not even clear to me whether it would do hardware accelerated and especially modern OpenGL. VirtualGL & XVFB require virtual displays - which seems to be quite a overhead for a server-side application supposed to be used by multiple users concurrently. So I am wondering what exactly you are doing here? Aside from probably having specific drivers etc. Are there ways that you would be willing to share, to build something similar in small with consumer NVIDIA hardware?
Ok, osmesa seems to be pure software… forget about that comment :)
So I’m wondering whether the architecture looks something like this:
- backend server with multiple gpus
- running a daemon managing & utilizing those GPUs with virtual displays/contexts - providing an remote API to render to FBOs and return an image
- frontend servers providing a remote API to submit rendering jobs and schedule them according to available backend daemons / GPUs