Given there is no way to know what your hardware manufacturer puts behind the SATA interface of your SSD, ZFS will need to just not assume identical blocks are

Given there is no way to know what your hardware manufacturer puts behind the SATA interface of your SSD, ZFS will need to just not assume identical blocks are stored independently. • a random salt is not required in the metadata, a sequential serial number would do, and is easier (you only need know that last, not all previous numbers). • with the right checksum algorithm you can remove the previous serial and update to the next serial with only minor computation on the checksum. • you do not need to copy the metadata in ram if you can tolerate doing the writes sequentially. (This is will be a speed loss since the drive can't get all of them at once and schedule them optimally. It might be a big loss if the drive is also receiving a lot of other write requests at the same time.)