All,

I've been pondering of late about how much material is required to saturate a given amount of available upload bandwidth and have come up with what I hope are few useful data points based on a months worth of logs made available to me. Naturally the volume of shared content required to generate a given amount of upload depends on what is being shared - it's an inevitable fact that illegal mp3s of the latest improbably perky teen sensation's songs - and videos claiming to show said perky teen copulating with a goat - will be downloaded more often than legal content. With this in mind I've gathered data points for three categories of material:

• Illegal content - mp3s of top-ten hits (of any nation) from the last two years.
• Legal unique content - Free material not commonly available on limewire (I used out-of-copyright ASCII texts)
• Legal mirrored content - Copes of (hopefully) legal files which commonly show up more than 20 sources in LimeWire 3.6.

I've settled upon 'Total Upload Average' > 80% of available upload capacity on a peer-node as a suitable definition of 'saturation'. Data from nodes younger than four hours was ignored. Servers with a firewall blocking incoming connections on non-approved ports were treated seperately from those on a 'raw' internet connection.

So if you're wanting to upload 8K/second of popular legal files (and are not behind a firewall) you've downloaded, set aside 161MB x 8 = 1.3GB of disk space. If you've 20GB of unique legal content shared from behind a firewall, expect it to consume 20 x 1024 / 1023 = 20KB/s of upload capacity.

Hardly startling stuff, but there are a few interesting points to note. Firstly, people trying to download 'illegal content' are able to make far better use of file shared from behind a firewall. This indicates they are far less likely to be behind a firewall themselves - one might conclude they are novice net users unaware of good networking practices. Secondly, demand for legitimate content exists. While less popular than 'illegal content' by a factor of 6 or 10, that over 10% of the demand for file sharing is driven by legit content (at least on the gnet) is a heartening sign. Thirdly, downloaders of files which are widely available on the gnet significantly favour non-firewalled sources - at least some gnutella clients are able to select a subset of 'easy' hosts to download from.