If you want to hibernate a system, you need swap because hibernating is
dumping the main memory onto the swap (after zipping it). That is why those
who hibernate their systems are often recommended to choose at least as much
swap as RAM.
A system that never hibernates had better have a (small) swap partition than
no swap at all. With no swap, the kernel kills a process as soon as the
system runs out of RAM. It may happen even if the system has a huge amount of
RAM, e.g., because of a memory leak triggered in some special conditions. The
kernel's choice of the process to kill is rather arbitrary and it may not
kill the faulty process.
With 1 GB of swap (for instance), the user of a desktop system will have time
to notice that their system has become slow, to save their work, and to
choose what application(s) to close to free some RAM. For server systems, the
intervention (after an alert sent by a monitoring system) may take longer and
such systems had better have more swap.
