When should I use VLSM instead of fixed-length subnetting?
Use VLSM whenever one network has subnets of different sizes. It assigns the smallest prefix that fits each requirement so you do not waste address space on a small VLAN.
Why must VLSM allocations be sorted largest-first?
Allocating biggest blocks first leaves the smaller ones to slot into the remaining space without overlap. If you allocate small blocks first they fragment the pool and large requests fail.
How does VLSM differ from CIDR aggregation?
VLSM is allocation (splitting a supernet into sub-subnets of varying sizes). CIDR aggregation is the reverse: combining contiguous subnets back into a shorter prefix for advertisement.
Can VLSM be applied to IPv6?
Technically yes, but in practice IPv6 standardises on /64 LANs, /56 site allocations and /48 organisation allocations. VLSM is rarely needed because the address space is so large.