Question

I wrote the following module which encodes a UUID to an arbitrary base:

http://pypi.python.org/pypi/shortuuid/

Now, this gets it down to 22 symbols with the default alphabet while preserving uniqueness, but I was wondering how many (/which) digits I could cut off while maximising the retained uniqueness.

Are all the digits of a UUID equally random/unique, or are some digits more random than others? For example, if the first few digits are a machine/application-specific identifier, then obviously they would be less random than the last few. I haven’t noticed anything like this in my experiments, but I want to be sure before I advise people on it.

Will truncating it to, say, 8 digits have 1/57^8 probability of a clash, or does is the probability not uniform on the digits?

Answer 1

Because of the way UUID’s are constructed, it very much depends on the version. And yes, some will be more random than others.
http://en.wikipedia.org/wiki/Uuid#Version_1_.28MAC_address.29

One way to hack around this is by taking a hash (i.e. sha256 for example) of the UUID. Those hashes should be distributed in a uniform way.

_{Do note that I haven’t done a really thorough analysis here. My answer should be in the ballpark but I give no guarantee that it’s completely correct.}