Animoto - Scaling Through Viral Growth

Animoto is a very neat Amazon-powered application. Built on top of Amazon EC2, S3, and SQS, the site allows you to upload a series of images. It then generates a unique, attractive, and entertaining music video using your own music or something selected from the royalty-free library on the site. Last week I spoke to a group of Computer Science and IT students at Utah Valley State College. Before leaving Seattle I spent some time downloading images from their athletics site. I then combined this with some Southern Surf Syndicate music from The Penetrators and ended up with this really nice video:

There's a lot going on in the background. After the images and the music have been uploaded, proprietary algorithms analyze them and then render the final video. This can take an appreciable amount of time and requires a considerable amount of computing power.

Animoto co-founder and CEO Brad Jefferson stopped by Amazon HQ for a quick visit on Thursday. Earlier in the week we had seen their EC2 usage grow substantially and I was interested in learning more. Brad explained that they had introduced the Animoto Videos Facebook application about a month earlier and that it had done pretty well, with about 25,000 users signing up over the course of the month, with steady, linear growth.

The reaction from the Facebook community was positive, so the folks at Animoto decided to step it up a notch.  They noticed that a significant portion of users who installed the app never made their first Animoto video — yet the application (as they themselves admit) relies heavily on the 'wow' factor of seeing your first Animoto video and wanting to share it with your friends.  On Monday the team made a subtle but important change to their application: they auto-created a user's first Animoto video.

That did the trick!

They had 25,000 members on Monday, 50,000 on Tuesday, and 250,000 on Thursday. Their EC2 usage grew as well. For the last month or so they had been using between 50 and 100 instances. On Tuesday their usage peaked at around 400, Wednesday it was 900, and then 3400 instances as of Friday morning. Here's a chart:

We are really happy to see Animoto succeed and to be able to help them to scale up their user base and their application so quickly. I'm fairly certain that it would be difficult for them to get their hands on nearly 3500 compute nodes so quickly in any other way.

-- Jeff;

Comments

[Frédéric Sidler]

Great to see that they succeed. Really cool web service.

I know that the rendering of video is time consuming. So is it possible to give numbers relative to the number of video to process instead of members. The number you give are huge, 3500 instances for 250'000 members. How many videos do they process or how many hours of video did they process during these three days (from 04/15 to 04/18).

[Nickolai Leschov]

Yes, the numbers seem huge to me. 3500 instances for 250'000 members - this should mean about 71 users per server? And, do I get it right - $250'000 to $2'000'000 a month for servers - that sounds big for me (i.e for a startup with no prospect of profit in sight, at least)

[Robert Folkerts]

@ Nickolai

The number of servers corresponds with the number of concurrent users. The bill to use that is by the hour, so you basically have to pay for the 'area under the curve', not the height of the curve. So, they can handle the 'flash mob' at 3,500 servers * $0.10/hour = 350 dollars an hour or $8,400/day (this could be a factor of 4 or 8 more if they used the larger servers). This is still a chunk of money by my standards, but it is remarkably cheap for nearly instant access to millions of dollars in servers and floor space at a 'four nines' data center.

[James Gardiner]

Wow, a cool way to burn money.

The internet is proving to be a great place where you can do just about anything. Strange how in US the VC's will give you money to do projects like this. I equate this to climbing Mt Everest. A great thing to do, but has no real return on investment.

If only I could convince the US, VC;s to pay for my next big Skiing holiday...

Now that would be cool too....

[Bernd Eckenfels]

This chart also says that they spent 30.000USD in two days... while it is nice to buy so much capacity on demand, it is also a question, if the "nifty" marketing idea which requires so many resources (I can imagine the load will settle at about 10kUSD/Day) works in the end. There is some advantage in beeing resource bound :)

Bernd