As a pure means of communication the LANL preprint server, which later became arxiv, made the journals in its core areas of theoretical physics obsolete basically overnight, so vastly superior was it in terms of speed of distribution and ease of access.

To me it’s one of the high points of the web and Paul Ginsparg is an absolute hero for setting it up; he reviews some of its history at https://arxiv.org/abs/1108.2700.

I think it’s an interesting social question why other fields have taken so much longer to adopt the arxiv model.