Abstract: The online advertisement is the most important model of Internet profit. It makes the computing advertising a hot point of current research. For the real-time requirement and the high accuracy requirement in the online advertising system, the difference between the implementation and efficiency is compared based on the two popular platforms of Hadoop and Spark. And then a linear model is proposed combined with the technology of Spark to be applied in the advertising system. Finally the linear model is optimized from the aspect of numerical characteristics, iteration and step size. The testing result shows that the accuracy has a 5% to 10% increase after the optimization.