The following is a very simple example:

nums = parallelize([1, 2, 3, 4]) 

If I just want to make an RDD out of a plain old Python list, I can call the parallelize() function in Spark. That will convert a list of stuff, in this case, just the numbers, 1, 2, 3, 4, into an RDD object called nums.

That is the simplest case of creating an RDD, just from a hard-coded list of stuff. That list could come from anywhere; it doesn't have to be hard-coded either, but that kind of defeats the purpose of big data. I mean, if I have to load the entire Dataset into memory before I can create an RDD from it, what's the point?