kmeans inital cluster selection

[Mark Snow]

I was looking through the kmeans code. As I recall, a good way to pick the 
inital cluster positions is to choose random data points. Is there an easy way 
to do 'randomly select N records' in map reduce?