Based on http://blog.boundary.com/2012/01/12/Flake-A-decentralized-k-ordered-id-generation-service-in-Erlang.html which is in turn based on Twitter's Snowflake project: http://engineering.twitter.com/2010/06/announcing-snowflake.html
Generates unique ids based on the current epoch time, current machine identity, and a counter. The result are mostly ordered unique ids that require no synchronization between machines.
javac -cp .:./lib/uuid-3.2.1-SNAPSHOT.jar:./lib/commons-codec-1.6.jar src/uniq/UniqueId.java
java -cp .:./lib/uuid-3.2.1-SNAPSHOT.jar:./lib/commons-codec-1.6.jar:src uniq.UniqueId
2011 MacBook Pro, 2.8Ghz i7
$ time java -cp .:./lib/uuid-3.2.1-SNAPSHOT.jar:./lib/commons-codec-1.6.jar:src uniq.UniqueId 10000000 > out
real 0m18.498s
user 0m5.662s
sys 0m12.973s
$ ls -lh out
-rw-r--r-- 1 mumrah staff 153M May 22 10:52 out
540 id/ms
Since each id is 128 bits, the I/O rate comes out to 8640000 bytes per second, or 8.24 MB/s. Compare this to 66.59 MB/s which is the speed that the same system can "yes > omgyes". I guess that order of magnitude difference is the price of safety/synchronization.
$ time java -cp .:./lib/uuid-3.2.1-SNAPSHOT.jar:./lib/commons-codec-1.6.jar:src uniq.UniqueId 10000000 > /dev/null
real 0m7.794s
user 0m4.688s
sys 0m3.253s
1280 id/ms
$ time java -cp .:./lib/uuid-3.2.1-SNAPSHOT.jar:./lib/commons-codec-1.6.jar:src uniq.UniqueId 10000000 > foo
real 0m18.051s
user 0m5.594s
sys 0m12.594s
554 id/ms