On 12/20/24 02:12, Bart Van Assche wrote: > On 12/18/24 10:00 PM, Christoph Hellwig wrote: >> If you rely on order. If you are doing O_APPEND-style I/O on zonefs or >> using a real file systems it's perfectly fine. > > Using zone append operations defeats two of the advantages of zoned > storage. One of the advantages of zoned storage is that filesystems have > control over the layout of files on flash memory with regular writes. Zone append still allows that: you can still chose the zone you write to. > That advantage is lost when using zone append operations because it is > allowed to reorder these operations. Another advantage that is lost is > the reduction in size of the FTL translation table. When using zone FTL ? Ar you talking about the device side FTL ? Zone append does not makes things worse in any way compared to regular writes. That completely depends on the device design and if (or not) zones are exactly mapped to erase blocks. > append operations, the filesystem has to store the offsets returned by > zone append operations somewhere. With regular writes this is not > necessary. That is a very misleading statement. The FS *always* records the write location of data, be it done with regular writes or with zone append. The difference is that given that zone appendd operations may be reordered, you may need a little more metadata to record the location of data if your writes endup not being processed in the same order as issued. But in my opinion, that is a slight disadvantage that can be completely ignored given the code simplification and performance advantage that zone append brings. -- Damien Le Moal Western Digital Research
