The dynamical properties of double-stranded DNA are studied in the framework of the Peyrard-Bishop-Dauxois model using Langevin dynamics. Our simulations are analyzed in terms of two probability functions describing coherently localized separations ("bubbles") of the double strand. We find that the resulting bubble distributions are more sharply peaked at the active sites than found in thermodynamically obtained distributions. Our analysis ascribes this to the fact that the bubble life-times significantly afects the distribution function. We find that certain base-pair sequences promote long-lived bubbles and we argue that this is due to a length scale competition between the nonlinearity and disorder present in the system.