Has anyone used new Compute VACF feature?

Dear LAMMPS users,

I am interested in calculating self-diffusivity of water in a porous solid structure using Velocity Autocorrelation Function.
I have used lamps_7Dec2013 with its newly added "Compute VACF” command to obtain VACF of water molecules in my system. "Fix Ave/time” has been used to output the data. Although, I can easily obtain self-diffusivity of target molecules in this system using their Mean Squared Displacement, the results from VACF are surprisingly incorrect, in such a way that the VACF values continuously fluctuate around zero even after 35 ns! (please see the figure attached for VACF data). I wonder if anyone has any suggestion in this regard? Also, I am unsure whether my approach in using fix Ave/time is the correct way of outputting VACF data?
Following you can find the input script I have used:


To get a diffusion coeff from the VACF you have to time integrate
the values, on a small timescale. Compute vacf does not do that
for you, you have to do it yourself. Compute vacf simply outputs
the VACF at different times. At long timescales the plot will simply
oscillate around 0.0 b/c the velocities are uncorrelated.