Unveiling interesting numbers about VLC downloads

Every year in September the VideoLAN community organize the Open Multimedia Conference also known as the VideoLAN Developers Days or for short VDD. This year it will take place in Berlin the 3rd and 4th of September. For this occasion I gathered some interesting data from the VideoLAN download servers.

First, let me start with a bit of context. VideoLAN is a (French) non-profit organization with 31 active members as of today. The non-profit acts as an umbrella for a lot of multimedia related projects like libVLC, x264, DVBlast, libbluray and more. But of course, the most known is also the one that eats up most our bandwidth: VLC media player.

On the infrastructure side we rely on amazingly nice peoples and organizations all over the world that donate some of their bandwidth for a good cause. We host the main replication server also known as the rsync master and all the mirrors use the rsync protocol to fetch updates multiple times a day.

When you’re heading up to the VideoLAN website to download VLC, a piece of software written in Go, called mirrorbits, will find in just a fraction of a second the closest mirror to you that is alive and serving the file you’re looking at. That way downloads are fast and reliable. Note: actually most of the numbers presented here were exported from the mirrobits database.

Currently we have 87 mirrors in 44 countries and the total size of our repository is 46 GB. During the past year we had no major incident and no downtime (thanks to the redundancy) so that’s pretty close to a 100% availability.

Just for the past month (August 2016), which is usually not the highest of the year in terms of traffic, 568 TB were downloaded from our mirrors. That’s an average of 1.7 Gbit/s for the month and we counted 23,794,472 downloads for the same period. These numbers do not include the mobile ports (Android, iOS, WinRT) or the Linux versions which are served by their respective platforms.

Let’s focus a bit on VLC. We only have (somewhat) accurate statistics for Windows (both 32 and 64-bit) and MacOS but it gives an interesting scale anyway: MacOS count approximately for 1/10 th of the total. The Windows 64-bit version is surprisingly low but that’s because the 32-bit one is the default choice (and works on both 32 and 64-bit architectures).

On this other chart you can compare the trend of the downloads for some of the recent versions of VLC1. The peak of a release usually includes the automatic updates that are started in the week that follows (to ensure there is no major regression). For example the 2.2.2 and 2.2.3 were quite short-lived releases which did not get pushed through auto-updates because of bugs.

Let’s finish with a fun fact. Few months ago I’ve setup a live downloads counter on the VideoLAN website (just below the big Download VLC button). It started as a fun experiment but stayed that way ever since. Behind the scene is a websocket that connects to our download server as soon as you load the page, and receives updates until you leave the page. Simple.

So that should give us an idea of how many users we have on our frontpage at any given time. After aggregating the results I got the improbable number of 985 simultaneously connected users on the videolan.org frontpage. We’re a popular website but still, 985 is a lot and it grows month after months… weird. At best we should have around 300 connections tops. But since we don’t disconnect the websockets after a while it means you can basically stay connected forever! And some of these connections are up since weeks. So what are all these people doing here you may ask. Well, the reason is probably that a lot of these users went for a quick visit and didn’t close their tab since! Amazing.

That’s all for today folks! I hope you liked this article. Don’t forget to follow me on Twitter and if you’re around Berlin this weekend, come and say hi!

  1. For more accurate VLC downloads statistics by version and refreshed every day, you can head to https://www.videolan.org/vlc/stats/downloads.html↩︎


comments powered by Disqus