We present the results of molecular dynamics computer simulations of a binary Lennard-Jones liquid confined between two parallel rough walls. These walls are realized by frozen amorphous configurations of the same liquid and therefore the structural properties of the confined fluid are identical to the ones of the bulk system. Hence, this setup allows us to study how the relaxation dynamics is affected by the pure effect of confinement, i.e., if structural changes are completely avoided. We find that the local relaxation dynamics is a strong function of z, the distance of the particles from the wall, and that close to the surface the typical relaxation times are orders of magnitude larger than the ones in the bulk. Because of the cooperative nature of the particle dynamics, the slow dynamics also affects the dynamics of the particles for large values of z. Using various empirical laws, we are able to parametrize accurately the z dependence of the generalized incoherent intermediate scattering function F-s(q,z,t) and also the spatial dependence of structural relaxation times. These laws allow us to determine various dynamical length scales and we find that their temperature dependence is compatible with an Arrhenius law. Furthermore, we find that, at low temperatures, time- and space-dependent correlation functions fulfill a generalized factorization property similar to the one predicted by mode-coupling theory for bulk systems. For thin films and/or at sufficiently low temperatures, we find that the relaxation dynamics is influenced by the two walls in a strongly nonlinear way in that the slowing down is much stronger than the one expected from the presence of only one confining wall. Finally, we study the average dynamics of all liquid particles and find that the data can be described very well by a superposition of two relaxation processes that have clearly separated time scales. Since this is in contrast with the result of our analysis of the local dynamics, we argue that a correct interpretation of experimental data can be rather difficult.