TV Ratings, NetBeans & Solaris
By user12601629 on Oct 10, 2006
The Network World article discusses the challenges now being faced in measuring TV viewership as more mechanisms are being used watch TV. The funny thing is that Nielsen, the company that does the TV ratings, is being stumped by people using services that should be easier to measure. They don't know how to deal with downloads and the net (talk about old school!). For example, downloads of TV episodes from the iTunes store are very specific and trackable (at least by someone with access to the right data), but Nielsen can't yet handle that. Where as the number of people watching or listening to an over the air broadcast is nearly impossible to measure, but Nielsen has methologies developed over decades for that. They're going to need to move into the 21st century or they're dead.
In the world of connected services for IT, data is easier to gather and richer in content. One interesting example of this is usage of Integrated Development Environments (IDEs). Companies like Evans and IDC make a living off Nielsen-like surveys that estimate the number of developers using differnent technologies. The conclusions of these surveys can be very hard to interpret -- and suspect at times. However, NetBeans (my favorite IDE) contains an embedded connected service called the Autoupdate Center. About once a week the IDE phones home to check for new modules. This is a useful service to the developer using the IDE, but also provides tremendous data to the developers of the tool. Using the back-end of this service it is possible to check a simple finger print on each ID and establish the number of unique IDEs that run each month. These kinds of measures are far more accurate than surveys. Here is some example usage data (note the huge growth!) the NetBeans team published a few months back.
In another example, users of Update Connection System and Hosted register the machines they want to attach to the network. That gives Sun an exact count of usage on the Update network (e.g. we registered about 18,000 new systems to access the network last quarter). I also know how many are x86 and how many are SPARC. Surprisingly, more are now non-SPARC (~60% Intel/AMD vs ~40% SPARC). That tells you something about Solaris x86/x64 adoption. I even know how many CPUs each machine has (the largest system attached to the Update network is a 144 CPU SPARC system in Canada). The team building our internal reporting system even set it up so I can view a given day's registrations with the location of each system plotted on a Google map! Now that's cool.
You can't do that with statistical extrapolations.