E is the energy scale at which we use our renormalized perturbation expansion ; it should be "close" to the energy of the real or virtual particles considered.
Let me explain:
you know that the non-renormalized perturbation series can be written down, but diverges for non-trivial values of the bare coupling. By rewriting the bare coupling as a function of the "actual" coupling (by inverting the functional relationship between the bare coupling and a "defining interaction" at a certain energy) we find finite expressions for the perturbation terms - this is the procedure of renormalization. However, it turns out that if you then want to use this expansion for interactions far away from the defining energy scale, you need to use a lot of terms (a lot of higher order Feynman diagrams), which is a pain. So you can now displace the "defining interaction energy" to another scale, but then of course you will find another relationship between the (redefined) actual coupling and the bare coupling ; or if you prefer, between the redefined actual coupling, and the original coupling at the original defining scale. This relationship is what you wrote down.
The nice thing is that if you work in the neighbourhood of the redefined scale with the redefined coupling, then you don't need as much diagrams to get to the same result (more or less).
In QCD it is a bit complicated, but let us take QED as an example:
there is a natural scale to define the EM coupling: that is for two electrons at rest at large distance, we want to find back Coulomb's law (with its coefficient e^2/4pi epsilon_0).
So the original defining scale is the rest mass of the electron in an electron-electron interaction. It is the original definition of alpha_0.
If you now use that alpha_0 to consider electron-positron collisions at 500 GeV, you need to sum quite some diagrams ; however, if you *redefine* alpha(500GeV) as what you would have for an electron-electron interaction at that energy, and you work it out as a function of alpha_0 (in order to do so you have to sum several diagrams of course), then an electron-positron interaction at 500 GeV (or 400 or 600 GeV for that matter) will already give you a good result when calculated at treelevel, with alpha(500GeV).
This is the "rescaling" of the coupling constant, and the resulting "running of the coupling constant".
cheers,
Patrick.