A Standard for Measuring and Reporting Downloads
By Gary Zellerbach on Jul 08, 2008
Mozilla and Firefox have garnered a lot of well-deserved publicity for their new Guinness record for most downloads in a day. This raises the question (again) of what constitutes a "download" and how is it measured? (I first touched on this subject a couple of years ago, writing "When is a download a download?".) I was glad to see in the above referenced BBC article, "The official figure was confirmed after logs from download servers were audited and checked to ensure duplicate and unfinished downloads were not counted." So that implies "attempted" downloads weren't counted, only completed. That's good, as I'm sure some stats we see reported don't make such distinctions.
So with Guiness in the mix now, and with all the focus on downloads in the free & open source software world, I propose it's time for an agreed upon standard on how the number of downloads is reported. (To be honest, I'm uncertain if this could ever be a "formal" standard and who would govern it -- please let me know if you have any suggestions.) For starters, here's what I propose:
- Simple case: One Product = one file
- This case means that the entire product is contained in a single downloadable file. If it is downloaded successfully, the customer can install and use the product.
- The suggested measurement is the one generally used -- the last byte of the file is delivered to the customer.
- Note this may not be 100% accurate. Threaded download managers can conceivably download the last byte of a file and then fail an earlier byte range before the entire download completes. But this is an edge case. It would also be extremely resource intensive to have to match up all the byte range requests in a threaded download scenario to "prove" that one user completed the download. Thus, I think the "de facto" standard of last byte delivered is acceptable.
- Complex case: One Product = multiple files
- First, make the distinction between required and optional files:
- Required files are required to install the product. There may be one or more required files as part of the download.
- For example, on Sun Download Center, there is an option to download Solaris OS as 5 CDs. If you do not download all five, you cannot install the product. Thus, all five are required files, and each one must be downloaded to install Solaris.
- Optional files are not required to install or run the product. Optional files may consist of "plug-ins," "add-ons," or "modules" that add functionality to the base program. Also included would be documentation, checksum files, additional language packages, and the like.
- The measurement of a completed download in this case is that all required files are downloaded by the same customer. (This can be a challenge to measure, but it can be done in most cases with a good degree of accuracy.)
- Optional files do not count and should never be reported as "product" downloads.
- In either case, "Attempted Downloads" should never be reported as a product download.
- For a single file product, an attempted download is when the user starts the download but does not receive the last byte of the file.
- For complex products, an attempt is when any individual required file is not completed and/or the customer does not successfully download all required files.