@lordZouga: Once will never be enough unless you're coming up with a breakthrough algorithm. Let me ask, have you tried implementing what you're thinking? If not try it!

₱®ÌИСΞ:


Hahaha lol...thnx for dat...I imagined the possibility of dat occuring in a HUGE file was near impossible so I guess my mind didn't go there...lemme fix it

Indeed, if N is big, and K relatively small, then this is pretty unlikely to happen. This was the subtle issue I alluded to in my first post. But it is pretty easy to fix (just reject a random number it is already in your list).

lordZOUGA:
it will take O(N)... This would be nice but your algorithm suggested doing it twice when in my opinion, once would be enough.
Yea most languages has internal buffer... C++ has the internal buffer object stream_buf that is created for every I/O operation

If you don't know N, I think that it is impossible to do it in one pass.

You have to know N to generate the K random row #s correctly. And the only way to know N is to pass through the file once.

OFDM?