OpenOffice.org is a fairly large project, available for many combinations of languages and platforms. Thus, we currently offer a broad range of files on our mirroring network that are quite large in size. Some of them are requested quite often, whereas some of them are old, legacy and only downloaded once in a while. Therefore, we need to establish a policy on how to mirror the files in an economical way.
Currently, we distribute all of them in two mirror modules: the "main" module and the "extended" module, whereas the "main" module is divided into several categories. The "main" module is mirrored by many mirrors, whereas the "extended" module - due to its size - is being mirrored only by a few ones. This hierarchy should not be changed. However, in order to shrink the size of all modules, we want to establish a new module called "archive", where we can store the mentioned old, legacy and rarely downloaded files and move them out of "main" or "extended" modules. The "archive" module is only mirrored on two or three mirrors to avoid too much traffic consumption. We currently DO NOT need new mirrors mirroring the archive set.
Location of the archive set
The archive set is available at:
Detailed structure of the modules
Roughly spoken, the modules should be functioning as follows:
The "main" set should only contain the last released version for each platform and language, including source code. For a transition time after each release, "main" should contain two builds until all download webpages have been adjusted.
The "extended" set should contain things like Beta Releases and Release Candidates (.../extended/<version number>) and developer builds (.../extended/developer/<milestone>) before they get final, as these files are requested quite often. "extended" is also the right choice for L10N, SDK, ISOs and additional dictionaries. Files that are only requested rarely should be put into "archive".
The "archive" module should contai old versions. However, as storing each and any version is quite senseless, and explodes disk space rapidly, only the last version of each code tree for the combination platform and language should be stored. Examples: 126.96.36.199, 1.1.5, eventually 2.4.1; but also things like 1.1.2 for LinuxPPC, which is the last version for this platform. Apart from its name, "archive" is also dedicated to builds that would normally fit into "extended" but are rarely downloaded.
And, last but not least: Experimental stuff, very old builds etc. should not be in the mirror at all. No one needs an unsupported OpenOffice.org version for a dead architecture on an unknown platform mirrored anywhere.