We study a comprehensive framework for sponsored search which incorporates advertiser budgets, query frequency forecasts, and pricing and ranking schemes. We propose a linear program for optimizing revenue (or the total value to advertisers) that has an exponential number of variables; however, we describe how it can be solved efficiently using column generation. The formulation is easily extendable to various levels of problem complexity, adaptable to dynamic environments, fast, and works well in terms of practical considerations. Simulations show significant improvements in revenue and efficiency.