How early should I start writing a paper for a conference?
Well, there are different approaches. Some people think "there is this conference coming, I need to write something for it", others think "I have this nice idea of research, let's see where I could present it". In the first case, you might need to start very early, since you need to do all the research, while in the second case, you might just need to start early...
Seriously though, it depends a lot of the content of the paper, the mechanical action of writing it (i.e., typing and formatting the text on a computer) is not what takes long.
Is it reasonable to write and do experiments at the same time?
Sure. Even if you can't finish your experiments by the deadline, it's very likely you can reuse a good part of what you've already written for another conference.
How many references should you check? one professor told me: check 200 use 30.
You don't check references because you should, but because they are relevant to your research problem. You should cite what you've read and what you think is important with respect to you work. You should cite the approaches that laid the basis for your own approaches, and explain how you build on them, which assumptions you are challenging. You should also cite the approaches that are similar to yours, and explain why yours is better.
There is no magic number: there are excellent papers with only 10 citations, and bad ones with 50 references.