Tuesday, May 3rd 2005

Firefox Downloads Without the Updates

Many people have lately pointed out that the Firefox download figure is inflated by people who download Firefox updates manually. (As opposed to updating from the options dialog within Firefox.)

Scott Kveton recently produced some rather sleek graphs which show the Firefox download progress since its launch back in November. This one in particular makes it possible to (fairly accurately) estimate the size of this claimed figure bloat: We simply compare the total area of the peaks around the dates of the 1.0.x releases with the area of the continuous flow of downloads prior to these releases. This estimation builds on one main assumption:

Looking at the download graph, this assumption doesn’t seem completely unrealistic, since the dl rate was pretty stable from the beginning of December until 1.0.1 was released in late February.

So, here are the results of my pixel counting:

1029 pixels were above the stable dl rate after 1.0.1 was released, and 2871 pixels can be considered legit downloads. This means that 26.4% or about 13,500,000 of the now 51,150,000 dls most likely were updates. That leaves 37,650,000 dls that basically can be considered “new people giving Firefox a try”.

Update: Asa D points out that Scott has in fact included the Firefox update downloads in these graphs. That means the numbers should be more like (1-0.264)x66,000,000, which is 48,6 million – pretty close to the 50 M number. That’s almost weird if you ask me – I’d have thought that more than 2-3 million people updated manually. Well, well…

No Responses to this post:

  1. Asa says:

    Two major problems here. One, the 50 Million figure doesn’t count the updates that are showing up in Scott’s graph. Downloads triggered by the update system are specifically excluded from the 50M number. The number including update-triggered downloads is considerably higher.The second problem here is that each of those updates also got us a lot more media exposure than we had the weeks before, and media exposure is our number one driver of downloads so there was very, very likely, a decent spike in new downloaders during those periods as well.It’s not like we have some major distribution program in force and the media exposure is some secondary driver. It’s _the_ driver of downloads so your note about assuming things remain steady when we get massive amounts of media coverage is a pretty large assumption. For example, I can actually see spikes in our daily downloads when we get a cnet story or when we have a popular campaign at spreadfirefox.com. During our updates, we’re seeing all of the tech press and some of the mainstream press linking to us and that drives new adoption more than anything else. Just as an aside, we also don’t count downloads from other hosts like download.com (they host builds for us rather than point at our sever, and have served millions of downloads of Firefox) and we don’t count downloads from people that find our ftp or http repositories directly (without going through a bouncer URL.)So, I guess my point is that if you’re going to try to actually measure new downloaders of Firefox, it’s insufficient to just try to peel off updates. You also have to go round up the various other sources of downloads and perform similar math on those. Our download count is a count of downloads. It’s not worth the effort to try to make it any more or less than what it is, a measure of people who came to our site and clicked on the “download Firefox” link. – A

  2. Asa says:

    So I may have gotten half of that wrong. If scott is counting just full downloads (and has already stripped out the update downloads) then the second part of my concern still holds.- A

  3. Asa says:

    Actually, I just checked with scott and it does indeed sound like he is including the update downloads in that graph. If you’re using his graph, then you’ll need to tack on another 15 million or so more downloads to the 51M total you were using. – A

  4. David Naylor says:

    Ok.I knew that the updates weren’t included in the 51 M figure. I didn’t know that Scott had included the updates in those graphs.I see your point about the media attention spikes. I realized the cut-off level I chose was probably a bit low, but I chose it so as to keep Opera/IE zealots from coming here and claiming I had chosen a too high cut-off…For the record, I wasn’t trying to calculate the total number of Firefox downloads ever, from all locations. I just thought I’d have a go att chopping the updates off that download figure… But since Scott includes the updates in that graph, my numbers obviously are a bit off…Actually, I was really trying to show that “hey, the download number isn’t as bloated as you people think it is”. :-)Thanks for the feedback anyways…

  5. David Naylor says:

    Oh – one more thing: I thought the number of FTP downloads was guesstimated and added into that download counter?

  6. Asa says:

    When you add in about 15 million more downloads to account for the downloads that show up in the graph but not in the download rss feed, I think the number comes out a good bit closer to 50 million.Also, no, we don’t estimate in anything from FTP/direct downloads. We’re only counting downloads that go through bouncer. In the Preview Release days, I was trying to estimate all this stuff but once we got the feed coming off of bouncer, I decided that was the easiest way to go and it’s simply a raw number (also missing the first two days when we did about 2.5 million downloads, I believe, becaues bouncer wasn’t in use immediately.)- A- A

  7. David Naylor says:

    Yes, after adding the 15 M updates into the calculation I get just over 48 M.OK. That’s why I thought the FTP dls were still being estimated. Nice to hear they aren’t now.I read about the missed dls in the first day over at Scotts blog.

  8. Scott Kveton says:

    For the record, as of 2:28PM PST on 5/5/2005 there have been 65,664,412 total downloads of Firefox (that includes new installs as well as updates).We don’t use ftp anymore because the errors that heavily used sites can generate are not very intuitive for end-users. Also, testing HTTP sites is a little easier for Bouncer.

  9. David Naylor says:

    OK. Thanks for the info.I have an idea: you could make updates a different colour in those graphs, so the tops of the bars would be a different shade or something… Well, it’s just a suggestion. Feel free to ignore it if you don’t want to implement it 😉