We assess the detectability of baryonic acoustic oscillations ( BAO ) in the power spectrum of galaxies using ultra large volume N-body simulations of the hierarchical clustering of dark matter and semi-analytical modelling of galaxy formation . A step-by-step illustration is given of the various effects ( nonlinear fluctuation growth , peculiar motions , nonlinear and scale dependent bias ) which systematically change the form of the galaxy power spectrum on large scales from the simple prediction of linear perturbation theory . Using a new method to extract the scale of the oscillations , we nevertheless find that the BAO approach gives an unbiased estimate of the sound horizon scale . Sampling variance remains the dominant source of error despite the huge volume of our simulation box ( = 2.41 h ^ { -3 } { Gpc } ^ { 3 } ) . We use our results to forecast the accuracy with which forthcoming surveys will be able to measure the sound horizon scale , s , and , hence constrain the dark energy equation of state parameter , w ( with simplifying assumptions and without marginalizing over the other cosmological parameters ) . Pan-STARRS could potentially yield a measurement with an accuracy of \Delta s / s = 0.5 - 0.7 \% ( corresponding to \Delta w \approx 2 - 3 \% ) , which is competitive with the proposed WFMOS survey ( \Delta s / s = 1 \% \Delta w \approx 4 \% ) . Achieving \Delta w \leq 1 \% using BAO alone is beyond any currently commissioned project and will require an all-sky spectroscopic survey , such as would be undertaken by the SPACE mission concept under proposal to ESA .