Portsnap usage statistics

The following graphs shows, for any given week, the number of days worth of updates to the ports tree fetched by each version of portsnap per day, averaged over the seven days of the week. This provides an estimate of portsnap usage which is biased only by the use of caching HTTP proxies (which will result in the numbers shown here being below the correct values).

To clarify what I'm measuring: By looking at the files fetched by portsnap, I can identify (for instance) that a system is updating its ports tree (or rather, a compressed snapshot thereof) from 1AM on August 2nd to 7AM on August 5th. This will count as 3.25 days of updates, and will be counted on August 5th (when the fetching was done). A system which updates once per day and a system which updates once per hour will both appear this way as 1.0 days of updates being performed per day.

Versions 0.9.x of portsnap are those distributed from this website and via the FreeBSD ports tree; versions 6.0 and 7.0 (and more to come) are copies of portsnap from the FreeBSD base system (and the version number refers to the version of FreeBSD).

I generate these statistics by gathering the access logs for portsnap.daemonology.net, portsnap1.freebsd.org, and portsnap2.freebsd.org, and running them through several perl scripts to extract, accumulate, and tabulate the values; I then use MetaPost (and mpgraph) to draw the graph, which I convert into a PNG file using peps.

Note that this is a stacked area graph: The height of each region corresponds to the number of users of that version -- for example, 0.9.3 appears partway up, but the green region is very small since that version of portsnap has very few users -- and the top boundary corresponds to the total usage of all versions combined. This sort of graph is useful when you want to convey the approximate relative size of several contributions over time along with their total.

A couple of blips on the graph which are worth pointing out:

The number of systems running portsnap is considerably lower during the week between Christmas and New Year's. As a result, there is a visible dip in statistics for the final weeks of 2005 and 2006, followed up an upwards spike as systems "catch up" in early January.
There is a sharp spike in the usage of Portsnap 0.9 shown in late March 2007; this is a result of version numbers not being logged for a few days on one of the Portsnap mirrors, and few or none of those updates were actually performed by systems running Portsnap 0.9.
There is a sharp spike in FreeBSD 6.1 usage shown in early May 2007 followed by a sharp drop in all the statistics. The spike is caused by a large network suddenly downloading many months of updates, while the sharp drop is caused by a combination of the xorg 7.2 ports freeze and a portsnap mirror not logging accesses properly for a few days.

Please send any comments about this to me at "cperciva" at this domain.