A single physical extent is the smallest unit of disk space that can be individually managed by LVM. So the extent size will determine the "granularity" of your disk space allocation.
The extent size is set when you create a VG, and all the physical volumes in a single VG will use the same extent size. Changing the extent size once the VG is in use will be difficult, so you should think about any special requirements the VG might have regarding the extent size at VG creation time.
If the available disk space is equal, using smaller extents will allow more fine-grained allocation if you need it, but conversely you may need more space for the metadata required to manage those extents. With basic LVs this is usually a non-issue, but might be important if you use LVM snapshots, thin volumes or other advanced functionality.
If you plan to have a multi-terabyte logical volume, you're unlikely to need to micromanage its size in increments of 1 Mb, so using a larger extent size can be beneficial as e.g. the snapshot metadata can be smaller.
If your disks have data alignment requirements, the PE size should ideally be an exact multiple of whatever the alignment requirement is, so you won't need to worry about performance issues caused by misalignment. Since the PE sizes are power-of-2 values anyway (just like common alignment requirements are), this is not usually an issue, unless you have e.g. SSDs with an unusual erase block size.
