On Thu, Dec 14, 2023 at 06:20:16PM -0800, Keith Busch wrote: > On Thu, Dec 14, 2023 at 06:03:11PM -0800, Jaegeuk Kim wrote: > > > > Okay, it seems there's first disconnect here, which fails to explain all the > > below gaps. Do you think the device supporting zone_append keeps LBAs inline > > with PBAs within a zone? E.g., LBA#n guarantees to map to PBA#n in a zone. > > If LBA order is exactly matching to the PBA order all the time, the mapping > > granularity is zone. Otherwise, it should be page. > > Yah, you're describing how 'zone append' works. Haha. I'm glad I read through all the answers as Damien and you already did the explaining. Note that in a complex SSD of course there can still be some remapping due to wear leveling, worn out blocks or similar policy decisions, but that usuall happens below the superblock level and the SSDs can be much smarter about the algorithms used for that as it knows all data will be written sequentially.
