Good day all- I'm still trying to gain critical understanding to allow me, with reasonable confidence, to use bcache / SSD caching of btrfs root file systems for debian and derivatives with mainline or backported 4.18.xx kernels. One significant roadblock: for decent performance and drive space utilization, bucket size on the bcache C device should be configured to be the same as the SSD's erase block size EBS; yet SSD mass marketers are on the whole notoriously tight lipped about publishing this line item in a public spec sheet, almost as though it might vary on a given model, from one production lot to another, or by firmware level. Just googling around forums and so forth turns up next to nothing. It seems certain kinds of benchmarking can help make an educated guess about EBS on the target device, but better to have that datum from a definitive source. Better yet, the industry ought to establish standards, with protocols in place for a device to report its geometry and attributes to the host OS at config time. Just as you should expect of any other mature electronic peripheral bringing in massive revenue. Another consideration. The "blocks" conversion tool to put a bcache superblock into a backing device (HDD) file system, has not made it into any debian derivative repository, afaict. They make you pull python3 source from github. Is it really possible to set up a cached device without such fs conversion, or is it integrated into "make-bcache -B"? If not, why would they incorporate a bcache module into distribution kernels without a necessary userland tool? Thoughts? (Rather not venture into bcachefs experiments since it means you roll all your own kernels) thanks in advance- TP [nonsubscriber]
