Hi Vihang
The simple truth is that GTEM cells have large meshing volumes (compared to wavelength) so unless you subdivide your free-space and excitation / metallisation regions into FEM and IE solver regions, the size of your FEM solution will be beyond the capability of all but the most powerful servers.
Have a look at the HFSS start-up screen, you'll notice an option to create an HFSS-IE design instead of an HFSS design. Using this option you can create your geometry and define boundaries (conforming to your geometry) to force separate solution of IE and FEM regions.
An approach to this simulation strategy has been published, though this refers to CST instead of HFSS:
**broken link removed**
Good luck!