Turbulence is a fundamental unsolved problem, at whose core are the multiscale processes that transfer, for example, energy across the inertial range of scales, or momentum across wall-bounded shear flows. Turbulence is also key to applications, from industrial design and energy generation to climate dynamics, where the worst uncertainties are often due to its modelling. Its practical computation and control have been hindered by empirical models and boundary conditions, in large part because of insufficient understanding of the multiscale transfer just mentioned. Direct simulations, without approximation, are expensive, but the past few years have seen the appearance of larger computers and reasonably-priced disks that allow, for the first time, the compilation of time-resolved data sets of canonical turbulent flows with high enough Reynolds numbers to be truly multiscale, as well as the possibility of performing conceptual experiments on them. The premise of this proposal is that those new capabilities should allow us to elucidate, once and for all, the physics underlying the multiscale transfer processes in turbulence in the next five years, especially in shear flows near walls. That will allow the formulation of more realistic engineering models, but the immediate goal of the proposal is to answer the fundamental questions that have resisted two centuries of attack by physicists and engineers. An important part of the work will involve adapting simulation codes to the new computer architectures expected in the next few years. Neither large-scale computing nor data mining are trivial activities, but our group has specialised in both during the past 20 years, particularly for the study of turbulence.