Choice of membrane system depends on what kind of lipids you want. POPC: PalmitoylOleoylPhosphoCholin, DPPC is DiPalmitoylPhosphatidylCholine. Both are phospholipids. If you want to simulate a biological membrane you need these kind of membrane systems. If you want to simulate SDS micelle, you need some SDS system.
Wikipedia has a nice article about the water models: https://en.wikipedia.org/wiki/Water_model
We choose boundary conditions such that it minimizes the volume. Cubic is larger than triclinic. Smaller volume means less number of atoms, thus, saves time in MD.
In regards to the water model used, as with any set of potentials for the same molecule, one potential will usually perform best at reproducing certain properties. If the computational expense for the water model is not an issue, you generally want to choose the model that reproduces your property of interest the best.
For example, surface tension might be more accurately reproduced with the TIP4P water model, even if other properties are not as well reproduced (note again that this is just an example, consult the literature if you want to know what reproduces surface tension the best). It all depends on what you are setting out to determine with the MD calculations.