Very large and geometrically complex scenes, exceeding millions of polygons and hundreds of objects, arise naturally in many areas of interactive computer graphics. Time-critical rendering of such scenes requires the ability to trade visual quality with speed. Previous work has shown that this can be done by representing individual scene components as multiresolution triangle meshes, and performing at each frame a convex constrained optimization to choose the mesh resolutions that maximize image quality while meeting timing constraints. The authors demonstrate that the nonlinear optimization problem with linear constraints associated to a large class of quality estimation heuristics is efficiently solved using an active-set strategy. By exploiting the problem structure, Lagrange multiplier estimates and equality-constrained problem solutions are computed in linear time. Results show that our algorithms and data structures provide low memory overhead, smooth level-of-detail control, and guarantee, within acceptable limits, a uniform, bounded frame rate even for widely changing viewing conditions. Implementation details are presented along with the results of tests for memory needs, algorithm timing, and efficacy