As stated in Chapter 11, page 461:

A good "eigenpackage" will provide separate routines

compute only the nth largest eigenvalues and corrersponding eigenvectors for the following desired calculations:

1) all eigenvalues and no eigenvectors
2)all eigenvalues and some corresponding eigenvectors
3)all eigenvalues and all corresponding eigenvectors

For my application, the matrix is real,symmetric,the size of whick is very lagre ,for example 3600*3600. In addition, only some largest eigenvalues and corresponding eigenvectors are needed. So I prefer the second situation above.
Would you like to give some clue hint to deal with the problem.
It will be pefect if source code is provided.

Thanks