Knowledge reduction algorithms based on rough set play an important role in KDD because of its advantage in dealing with uncertain data. However, it is hard for classical rough set knowledge reduction algorithms to deal with huge data sets. A structure of Class Distribution List (CDL) is presented in this paper to express the distribution of all attribute values in the whole sample space. With database technology, a CDL can be generated through classifying the original data sets. Then, a group of rough-set-based knowledge reduction algorithms are revised using CDL. This method can process huge data sets directly. As a framework, CDL method can also be used in other rough set algorithms to improve their scalability without decreasing their accuracy. Efficiency of our algorithms is proved by simulation experiments.