One of the issues I imagine that could impact you in a few years is that none of the elements of Bionode (that I observed) are “server class” and thus subject to consumer grade lifespans. If you don’t have spares of identical hardware elements, you risk not being able to “plug in replace” and keep going. Or have you virtualized everything (Docker or equivalent) so that the actual hardware is commodity?
Oh definitely. This is a development system and conceptual work in progress that is giving me immediate utility, but if elevated to the level of a deliverable product, there would certainly be some other choices. While going through all those learning curves, I'm taking advantage of tools that I can afford and integrate. It's a long arc.
But to answer your other question, yes, some of the tools are in containers (part of the challenge is that TrueNAS pivoted from Kubernetes to Docker before my time with it, and that leads to occasional confusion, including such mundane things as AI giving completely wrong advice!). In any case, yes, that is the architectural evolutionary path within context of reasonably fast feedback with an asymptotically minimal development team of a little less than one. Machine learning component for Immich in the NAS, for example, is over in the MS-01 using its adorable little GPU, and I have visions of a more powerful standalone LLM host that isn't weirdly encased in a Windows environment that always confuses me.
One of the issues I imagine that could impact you in a few years is that none of the elements of Bionode (that I observed) are “server class” and thus subject to consumer grade lifespans. If you don’t have spares of identical hardware elements, you risk not being able to “plug in replace” and keep going. Or have you virtualized everything (Docker or equivalent) so that the actual hardware is commodity?
Oh definitely. This is a development system and conceptual work in progress that is giving me immediate utility, but if elevated to the level of a deliverable product, there would certainly be some other choices. While going through all those learning curves, I'm taking advantage of tools that I can afford and integrate. It's a long arc.
But to answer your other question, yes, some of the tools are in containers (part of the challenge is that TrueNAS pivoted from Kubernetes to Docker before my time with it, and that leads to occasional confusion, including such mundane things as AI giving completely wrong advice!). In any case, yes, that is the architectural evolutionary path within context of reasonably fast feedback with an asymptotically minimal development team of a little less than one. Machine learning component for Immich in the NAS, for example, is over in the MS-01 using its adorable little GPU, and I have visions of a more powerful standalone LLM host that isn't weirdly encased in a Windows environment that always confuses me.