The CPAN is the result of many volunteers working together, many of whom were originally operating their own little (or big) Perl FTP sites back before that Web thing came along. They got coordinated on their perl-packrats mailing list in late 1993 and decided that disk was getting cheap enough that the same information should be replicated on all sites, rather than having specialization on each site. The idea took about a year to ferment, and Jarkko Hietaniemi established his Finnish FTP site as the CPAN mothership, from which all other mirrors could draw their daily or hourly updates.
Part of the work involved rearranging and organizing the separate archives. Places were established for Perl binaries for non-Unix architectures, scripts, and Perl's source code itself. But the modules portion has come to be the largest and most interesting part of the CPAN.
The modules in the CPAN are organized as a symbolic-link tree in hierarchical functional categories, pointing to author directories where the actual files are located. The modules area also contains indicies that are generally in easy-to-parse-with-Perl formats, such as the Data::Dumper output for the detailed module index. Of course, these indicies are all derived automatically from databases at the master server using Perl programs. (Often, the mirroring of the CPAN from one server to another is done with a now-ancient Perl program called mirror.pl.)
From its small start of a few mirror machines, the CPAN has now grown to over 200 public archives in all corners of the Net, all churning away updating at least daily, sometimes as frequently as hourly. No matter where you are in the world, you can find a nearby CPAN mirror from which to pull the latest goodies.