In order to improve the performance of Deep Web crawler in discovering and searching data sources interfaces

a new method was raised to parallel processing the mass data within the Deep Web compromising MapReduce program-ming model and virtualization technology.The new crawling architecture was designed with three producers

the link classified MapReduce

the page classified MapReduce and the form classified MapReduce.Server virtualization was adopted to simulate the cluster environment in order to test the performance.Experiment results indicate that this method is capable for large-scale data parallel computing

can improve the crawling efficiency and avoid wasteful expenditure

which prove the feasibility of applying cloudy technologies into Deep Web data mining field.