Best solution: Short Random String?
Upon further reflection and based on the replies so far, assuming your database returns an error on insertion of a duplicate indexed unique value (which I think all should), I am actually thinking the most efficient solution is an 8 digit or so random alphanumeric string using something like this:
https://github.com/puid/Elixir
62^8 = 2.18340105585e+14
If your database is automatically returning collisions as errors and you automatically regenerate the ID and try again (as I think any safe system should regardless), then even if these collisions happen once a week or once a day you are still saving major efficiency overall.
If you go to an 12 digit random case in-sensitive alphanumeric string (in case database primary key is not case sensitive) this is only 12 bytes (1/3 of the UUID) and is on the same order of magnitude of value count as the positive range of a 64 bit Int UUID.
Why “Efficiency”
When I speak of efficiency, I refer to data storage costs as well as data transit. For example, I would in fact be doing some multiplayer gameplay where I would want to convey positions/rotations/appearance/etc of users with their ID’s on a continuous updating basis.
In terms of data transit, I believe string ID’s of any random nature do not compress well as they are random data already. I gave an example where you can be conveying 36 KB/s of data transit per user just on UUID’s (100 users * 36 bytes * 10 fps).
You can cut your data transit cost by 78% by using 8 digit strings in the JSON by contrast in this situation and avoid having to map users to shorter temporary ID’s on the fly.
Most importantly, if the database does return collisions as errors, and you can handle this and they are rare, why store bulky ID’s for no benefit?
