I have solve linear equation system A.x=b, where number of equations more than number of variables (in terms of LMS).

I find a system solution x via QR decomposition: A=Q.R
using qrdcmp().

Unfortunately, a linear system sensitivity is needed in my application, also.

A simple way, it is calculating of symmetric info-matrix
constructed as Transpose[R].R
and than finding it's eigenvalues and eigenvectors
using tqli() and tred2().

However, I will be appriciating for any referencies on SVD methods specially designed for triangular matrices:
R=Transpose[U].W.V

Evgeny