A little background to this migration: We had a large enterprise-scale implementation with dozens of JEE projects of all sizes (from a mouse to the ocean liner - and NO! I am not talking about the Titanic!), and at least 200 active developers accessing the system simultaneously during the day. Most projects are using Maven. To help teams to make better and faster builds we offered a complete infrastructure for continuous integration: a big server (Dual QuadCore Xeon with 16GB RAM running Red Hat Enterprise Linux 5.1), hosting an Atlassian Bamboo continuous integration server with (approximately): 110 (real time) continuous integration builds, 40 (daily) maven sites builds, 20 (daily) Sonar builds. It also hosted a Maven repository server (Archiva 1.1.1 before migrating), which stored a little less than a dozen local repositories and provided a proxy and cache for twenty external repositories (for almost 400GB of data). The majority of that space was used by the repository of internal snapshots (200GB) and the repository of internal releases (150GB). To understand these sizes you must know that some projects may produce (for good or bad reasons) some EARs up to 100MB in size. By adding the sizes of JARs and WARs that compose these EARs, each deployment quickly grows to take up a lot of space.
Read the rest here: http://www.sonatype.com/people/2009/07/from-apache-archiva-to-sonatype-nexus/