DFS Overview

Introduction

Each phase of Universal Communications will require maintanence of central archives and replication of selections to remote locations, including replication through a Delay and Disruption Tolerant Network (DTN) that may include postal/courier and broadcast links. Updates to metadata, such as catalogue information concerning available podcast channels and episodes should be maintained through a directory service which points to permanent locations within a Distributed File System accessible from all locations (either online or after a delay).

Various different aspects have similar requirements eg software repositories, OLPC library and podcasting. The usual solution for replication is to use the standard system administration tool rsync to maintain "mirrors" of directory trees on multiple servers. This approach only works with full internet connectivity and does not scale well. It is better to use a proper DFS designed to work in a DTN environment from the start, and enhance it as  required for later phases.

Tierstore

A DFS suitable for replication via DTN may be Tierstore developed by technology and infrastructure for emerging regions for the specific needs of community radio stations in the least developed countries. A paper from February 2008 looks very promising. But only alpha software from 2007 appears to be publicly available.

The design supports subscriptions to subtrees of a unix file system, including hard links. Whether or not it is suitable for more advanced applications mentioned in the above paper, that approach seems sound for supporting podcasting channels. In particular an audio track could be made available in multiple channels for retention over various periods simply by hard linking to appropriate directories, so that different subscribers would receive  and retain the track under different channel names with different local available storage space for retention, but the audio file itself would only be replicated once. Likewise a file submitted from downstream to a subtree only propagated upstream could be reviewed upstream and hard linked to one or more downstream channels on acceptance. It should also be straight forward to "broadcast" a large collection, including hard links, via ordinary DVD media.