You might consider then getting four 40gbps nVME enclosures, and then RAIDing multiple together (e.g. in a big stripe, you could get 160gbps throughput, only limited by # physical interfaces). Each slice could be +TBs.
Obviously increases your failure rate, but if you're constantly updating the same models (and not creating your own) you don't really need redundancy.
Comments
10gbps is slow enough to be annoying when you're loading a 200GB model, unfortunately.
You might consider then getting four 40gbps nVME enclosures, and then RAIDing multiple together (e.g. in a big stripe, you could get 160gbps throughput, only limited by # physical interfaces). Each slice could be +TBs.
Obviously increases your failure rate, but if you're constantly updating the same models (and not creating your own) you don't really need redundancy.