Hi Kalaiselvi,
I posted something similar to this on the Radiation Oncology Blog some time ago. I'll repost it here:
Superposition/convolution: The primary beam attenuation is determined in each voxel, and then a dose kernel is applied to each voxel to model the scatter.
Pencil beam: The beam is divided into small parts (pencil beams) that enter the surface of the volume. For each beam, a kernel that represents the dose distribution arising from it as it passes through the volume is applied (and adjusted for inhomogeneity etc). The dose distribution for the entire beam is formed by summing the dose distribution for each pencil beam kernel.
Less kernels = faster processing, therefore the pencil beam method would be faster than superposition/convolution.
Less kernels = less accuracy
In terms of anisotropy, I'm not really up to the formulas and how this is calculated, but I believe it is individual for each type of source and that the computer models account for this.