This article (in older forms) has been floating around for a very, very long time on Dave Johnston's website. I remember it inspiring me to work harder on my own attempts at level design but also that there's a top tier skill (and level of commitment) that I eventually realized to never reach. It's mind-boggling to see how many considerations go into making a great (not good, great) map, even in much simpler times before true photorealism. Yet even then you can never predict how well it will work out in the end. It's a really interesting discipline between game design, architecture and understanding the technical details of the underlying engine.
To be honest, I often feel like maps in competitive games become popular for seemingly no reason at all. As a long-time CS player, dust and dust2 have always felt rather stale to me. I've always found the dynamic, multi-level combat of maps like assault and office to be more fun. Fortunately, I've been hearing a lot more people piping up in favor of traditionally unbalanced maps like nuke these days!
(What's annoying is when people try to attribute a map's popularity to its design. I've heard some people say that fy_iceworld is a brilliantly designed map. Seriously.)
Iceworld is brilliant design relative to what it set out to do - strip away the context and provide an arena to practice your aim. The only thing it doesn't do is provide elevation (other than if you or someone else jumps on a center wall, but that's asking for trouble), but other than that you've pretty much got everything else.
Exactly. It's amazing such a beloved and iconic example of game design art began as merely an attempt to recreate a screenshot. No lengthy design documents, no big ideas, no gray box test "notes", no countless iterations. Great things can seem to just come into existence.