Some properties of the Arnoldi based methods for linear ill-posed problems