If you want to debug kernel crashes, then you need at least swap==RAM.
Beyond that, you need to think about how many simultaneous connections will be handled, how many web sessions there will be, how many ... etc. Figure out how much RAM that will all take, add 10-20%, and then make sure that RAM+swap covers that number.
Or, if you have several gig of RAM, just create a 1-2 GB swap, and monitor swap usage. If it gets above a couple hundres MB, then you know to add more RAM to the server.
Nowadays, I view swap usage more as a hint for when to add RAM, then as a safety-net for running out of RAM. I've yet to find a server that would burn through RAM faster than our notification software could alert us to an issue.