>>>- My first thought is we'll eventually want to migrate the existing BLOB data out to the FileStream and work just from that column. Otherwise we'd have to search both the existing and new columns for stored files and UNION the results. If that's the case, any tips about the migration process? My understanding is no users have huge amounts of data in BLOBs but it could be multiple gigabytes (order of magnitude)
>>
>>
>>If HW capabilities are sufficient, correct a backup DB and only synch data changes done after backup creation. You have ample time to make that process error free.
>
>I don't understand the concept - could you please elaborate?

Implement having an audit log for all data changes in the DB.
Create a full backup of original DB at specific point X in time.
Create a play DB from backup media.
Run scripts
to alter DB and
move blob data in play DB to filestream.
Run script on audit logs to create script file mirroring all changes done to org in play up to time Y,
or
create an incremental backup of changes after X in org, import those changes in play and re-run the blob move data script filtered for dtChange>=X
run alter script to eliminate blob data fields
verify that play is functionallly equivalent to org, rinse and repeat if not (with better estimates for "good" X and Y).