Neither of those links are particularly useful to someone wanting to know how to take an algorithm and
determine its runtime efficiency. They're both great for mathematicians though.
Big O is used to compute the worstcase "runtime efficiency" of an algorithm. I quote the "runtime efficiency"
because there isn't a single definition of it. For example, I might want to put an upper bound on the number
of memory accesses a particular function performs. Or I might want to put an upper bound on the amount
of memory a function uses.
In academia, usually you are asked to compute the number of times an algorithm loops or the number
of comparisons a sorting algorithm will perform. In any case, usually the input to the function is an
array (list, whatever) of N elements.
There is no magic to it. You simply have to count whatever it is you want to count  memory accesses,
loops/iterations, memory used, whatever. Since the input to the function is an arbitrary number of
elements (I used N above), usually you can't arrive at an exact number, but you can come up with
a formula written in terms of N.
As an example, compute the number of loops (ie, number of times "new" is invoked below) in the following algorithm:
1 2 3 4 5

void foo( int* array, size_t N ) {
for( size_t i = 0; i < N; ++i ) // Outer loop
for( size_t j = 0; j < i; ++j ) // Inner loop
new int[ 10 ];
}

Let's count the number of times new executes.
The first time we arrive at Inner loop, i == 0. j starts at zero, and j is not less than i, so the loop executes 0 times.
The second time we arrive at Inner loop, i == 1. j starts at zero, so the new executes. j is then incremented, and now j == 1, which is not less than i, so the new is not called again this time.
The third time we arrive at Inner loop, i == 2. You can easily figure out that new will be called twice this time.
The fourth time... you should be able to see the pattern.
1st time == 0 calls;
2nd time == 1;
3rd time == 2;
4th time == 3;
Nth time == N1
So the total number of times new is called is 1 + 2 + 3 + ... + N  1.
You should know that the sum of the first k integers is k(k+1)/2, so substituting N1 for k, you get
(N1)(N)/2 == ( N^2  N ) / 2.
Now according to the rules of bigO notation, O(cN^2) == O(N^2) where c is a constant, so
we can immediately drop the 1/2 factor and make it N^2  N. Again according to the rules
of bigO notation, if f(x) = g(x) + h(x), then O(f(x)) = g(x) if g(x) grows faster than h(x) or h(x)
if h(x) grows faster. In our case, N^2 grows faster than N, so our algorithm is O(N^2).