Summary

  • In March 2004, Debian described using popularity-contest reports to order packages across installation CDs, alongside other selection criteria for parts of the first two discs.
  • The survey inferred use from file timestamps. Recent upgrades, missing suitable files and disabled access-time recording each changed what the numbers could say.

Thirteen CDs make a fairly imposing welcome mat. In March 2004, Petter Reinholdtsen described the Debian release then being assembled as about 13,500 packages across 13 binary discs. The immediate problem was a seating plan: which software should get a place near the front?

His appeal to users explained the arrangement. Some packages on the first and second CDs were selected under other criteria; the rest were ordered using usage reported by popularity-contest, usually called Popcon. Computers that joined the survey could help arrange what the next installer encountered. A little information travelling upstream would shape the distribution travelling back down.

The hard drive's weekly postcard

The documentation shipped with version 1.28 in May 2005 describes a weekly email containing package information. Under the default cron schedule documented there, the job ran on Sunday at 6:47am. Debian had found a way to conduct market research before breakfast.

For each package, the program looked for a relevant recently accessed file and reported its timestamps. That made an installation list more useful: a machine could have software sitting on its disk for months while repeatedly using something else. The timestamp offered a trace of activity that mere possession lacked.

The clock also needed interpretation. Installing or upgrading a package could leave the access and change times close together, so the program gave that situation its own label. Packages without suitable program files went into another category. If /usr was mounted with access-time recording disabled, the FAQ said the report supplied the installed list without usage information. A quiet clock could be a configuration choice.

The distinction was visible in a report dated 20 January 2004. It offered separate views of installations, regular use, older use, recent upgrades and insufficient timestamp information. The page counted 1,707 submissions for that view. Those separate columns were doing useful editorial work: they kept several different stories from being squeezed into one popularity number.

There was also the question of software that worked behind the scenes. The version 1.28 README pointed out that a package looking old could still be required by another package in active use. A library can earn its keep without attracting a reader's personal affection. Removing it to tidy the rankings would be an ambitious way to break a favourite program.

Popcon connected a practical publishing decision to a practical observation. Earlier placement could make an installation disc more useful. The observation came from machines whose operators opted in and whose reports arrived, filtered through the way files recorded access. Debian's CD makers still had to decide how that evidence should affect the seating plan.