On 07/12/2020 20:27, Javier González wrote: > Good point. We can share some performance data on how Simple Copy scales > in terms of bw / latency and the CPU usage. Do you have anything else in > mind? > With an emulation in the kernel, we could make the usd "backend" implementation configurable. So if the emulation is faster, users can select the emulation, if the device is faster then the device. Kind of what the crypto and raid code do as well. I'm really interested in this work, as BTRFS relocation/balance will have potential benefits, but we need to get it right.
