craig@NNSC.NSF.NET (Craig Partridge) (12/12/90)
Jon Postel recently reposted a note to namedroppers from Don Wells, which asked, among other things, how ComputerWorld estimated 3 million users. I don't know how they did it, but I know how most other publications have done it (they've asked me) and thought I'd give a brief explanation. A while ago, I sat down with a couple of folks to design a poll to estimate the number of users on the Internet. In practice, none of us turned out to have the spare cycles to do the full poll, but we did do a quick and dirty preliminary survey. The goal of that survey was to try to get some handle on the distribution of the number of users per machine, and to see if we could get some correlation between information in the HINFO RR fields and the number of users. (This would allow us to do several small polls of different HINFO-selected groups and then combine the data with pretty high accuracy). One surprise from that survey was that the mean number of users per host appears to be closer to ten than to one. We had thought that single user workstations would swamp mainframes -- that apparently wasn't true. Thus, a very crude estimate of number of users is to multiply the number of hosts by the mean number of users. I usually use 5 as the mean number of users, and the host count from the treewalker program at the NIC. Craig PS: Re "users". One can easily get into debates about double-counting. We asked for a count of people who actually received e-mail on the machine (i.e. weren't forwarded). That gives a pretty good estimate.