when you are finding the volume of a solid of revolution, you are simply cutting up the solid into infinitely many differential pieces. To find the volume you simply sum the areas of each differential piece, by taking the integral.
The key to figuring out which method to use is by picturing it in your mind. Suppose you had a sine function from [0, pi] revolved around the x-axis. To find the area below the curve, you would divide the area into differential, vertical rectangles going from 0 to pi. Well now, you're not finding the area; you're finding the volume. So, just as the area is revolved around the x-axis to form a solid with a volume, why not revolve each differential rectangle around to form a circle (a disc), and then sum up all the circles to get the volume of the solid? You see where I'm going here? Each circle would be perpendicular to the x-axis and you'd be summing them up in the x-direction.
Now let's think about it a different way. Instead of dividing the area up into differential, vertical rectangles, why not divide it up into differential horizontal rectangles, and then summing the horizontal rectangles from the absolute max of the function in said interval, down to the x axis? But hey, we're finding volume... so we revolve said rectangles around. Now, can you picture what would happen? You'd get concentric shells, parallel to the x-axis. And to find the volume you'd have to find of the surface area of each shell, and sum them up in the y-direction.
Sorta a long post, but I've always found it easier to visualize this to conceptualize why things are, rather than straight memorization. But yes, shells are used parallel to the axis of revolution, and discs are used perpendicular to the axis of revolution.