Until 1999, the standard way of modelling the Internet was to use randomly generated graphs, in which routers were represented by points and the links between them by lines. But it turns out that such random graphs are a poor approximation because they miss two important features. The first is that links in the net are preferentially attached: a router that has many links to it is likely to attract still more links; one that does not, will not. The second is that the Internet has more clusters of connected points than random graphs do. These two properties give the Internet a topology that is scale-freein other words, small bits of it, when suitably magnified, resemble the whole.