By linear programming we can achieve the best result(maximum profit or minimum cost) in particular conditions. One of the most efficient tools for solving linear problem is simplex method. Simplex algorithm is not efficient for large scale programs. Since real problem (because of existence of many variables and constraints) designs in large scale, we need to an algorithm that converges to solution fast. These programs exist in data mining, machine learning, and Newton algorithm that converts the linear program to an unconstrained nonlinear program. In this paper we develop one of the method of recent researches for more structure of linear programming by using of decrease iterative algorithm and based on Newton method and penalty functions we present an efficient method for solving large scale linear programs. For this we prove the method and then test it with random numerical problems. The result show that proposed method is efficient enough and has sufficient accuracy and speed. Therefore we have presented a fast Newton lgorithm, for solving a class of linear programs and computational testing on test problems with a very large number of constraints and a moderate number of variables demonstrate the effectiveness of the proposed method in comparison with a state-of-the-art linear programming solver for this class of linear programs with three indices: primal solvability, dual solvability and norm of difference of primal and dual objective function.