By structuring it based on task id, he is able to generate id's that are guaranteed unique across an entire map-reduce job. Very large MR jobs would just touch the birthday zone for random integers so he figured it was better to be a bit more structured.
On Sun, Feb 21, 2010 at 4:50 AM, Sean Owen <sro...@gmail.com> wrote: > What's the bit at the end? looks like pseudo-random number generation, > so just use a Random? > -- Ted Dunning, CTO DeepDyve
