This is probably best done through apache with a combination of the mod_log_config module and mod_logio module that is available in Apache 2.0.x.

You haven't mentioned how you will identify your clients, but it will probably be done with a REFERER header (easily spoofed by end users) or a cookie or query parameter. Either way, you can use the above modules to generate a custom log to track actual bandwidth used.

Something like the following:

LogFormat "%V %a %s %I %O %{Referer}i %q" trackbandwidth
CustomLog /var/log/bandwidth trackbandwidth
[download]

You will have to choose what you save in the log file, whether it is the cookie headers, or the query string (this will depend on how you plan to identify your clients). The important entries above are the %I and %O which contain the actual number of bytes sent and received for this request (this includes the headers and the content of the request). I am pretty sure that this will take into consideration a user stopping the request part way through the download as well, but I would suggest you test that before taking my word for it.

Then just write a perl script to parse this log file and generate a full report of the bandwidth used per client.

Stopped requests are a real concern because of the way download accelerators like Go!Zilla, FlashGet, and GetRight often work. The more uncouth ones will request a file multiple times at the same time at different Range offsets and then drop connections once they get an overlap - so make sure the %O in mod_logio really works. What a great Apache module! I wish had been available back in 1999-2000, when I was doing this sort of thing. Actually, I was doing the reverse - trying to regulate a free geocities-type site (and prevent residents from using more than their fair share of bandwidth). I ended up counting each partial request at the full size of the file, primarily punishing the warez and pr0n crowd (who tend to use download accelerators the most), which was fine with me.

To keep check of the bandwidth, you might want to look at something like IPTraf. This piece of software was used on a campus I know of, to keep track of the bandwidth usage of the students, and worked reasonably well, IIRC.

HTH

Do you need any help?

I trade code for links ;)

Owen (aka 'ok' aka 'spentrent')