Hacker News Reader

Comment by leaveyou | Hacker News Reader

I don't know what infrastructure youtube has, but I work in the same building with another known "youtube clone" from Europe and they are spread on 7-8 floors and have a lot of remote infrastructure. (serious question) How can you realistically compete in this area with a basic youtube clone ? From my point of view, the infrastructure and operational costs dwarf the development costs.

View on HN

They're the clear market winner but some companies might need an internal youtube clone or might need to adapt a similar system that they would find his useful for. But hey if it has enough bells and whistles it could be the facebook to myspace.

The nice thing (at least from my admittedly-fairly-ad-hoc experience) about video processing is that it scales fairly linearly, at least for the basic features.

Things like building a scalable transcoding cluster can be made fairly simple ("just add more boxes") using a decent message-queue service. From there, everything is pretty simple HTTP.

Of course, this is only the very barebones stuff, and covering all of Youtube's features exhaustively would be quite difficult, and requires some good Machine Learning work.

Partly related anecdote: I did some contract work some years ago for a startup (Tellytopia, later acquired by Concurrent Systems) that was into the video over IP space. Video transcoding was a part of the project, but I didn't work on that, which was in C++. I worked on some of the surrounding parts, in Python with XML-RPC, talking to Java, across machines. I remember that all or most functions in the app were XML-RPC-enabled, so that they could be located in different machines, which must have helped somewhat with scaling.

I work on data science. I also had an idea around youtube. Want to have a chat?

Sure, you can email me at [email-redacted] (yes, really).

Reply on news.ycombinator.com