What I did (pre-Internet!) was to skim each of the many journals relevant to my field carried by the library when they came out, reading the papers that seemed relevant. If highly relevant, I'd photocopy them, scribbled complete details on the reference (journal, volume, number, pages, month, ...) on the back of the copy, wrote up a short (2 to 5 lines) summary into my "survey" chapter and added it to my bibliography database. If something relevant was in the references, I'd chase them down and applied the same technique. This survey, after pruning, became my "State of the art" chapter.
Writing a short summary forces to really understand the paper, having the summary helps finding the paper if some further details are needed later, the bibliography database will be invaluable (not just now, also later).
Today I do mostly the same, just that I keep electronic copies of the papers, and add a custom field to my BibTeX entries to keep the URL (which might point at some preprint, a technical report that gets summarized, or even not-so-legal sources that can't be mentioned officially).