3 ms·

Ive always worked at a much higher level than infra, and I used to be similarly dismissive about the difficulty of infrastructure at scale until I had some expo

Ive always worked at a much higher level than infra, and I used to be similarly dismissive about the difficulty of infrastructure at scale until I had some exposure to it; You'd be surprised at how much talent and effort and money goes into delivering the volume of content that YouTube does at the reliability it does. The fact that I subconsciously always pick YouTube video search results links over the alternatives has a lot to do with the fact that it's apparently difficult enough for every other entity to deliver video content reliably (with far, far lower volume).

sam0x17 7y ago
Right but all of that auto scaling has been essentially donated to the community in the form of google cloud platform and AWS. Now we have cloud functions / lambdas (that are capable of running ffmpeg), Kubernetes, S3, Docker, and a whole slew of things that make this task fairly trivial compared to how it was in say 2005. There also isn't any rule saying you have to do your video conversions server side. Imagine the cost efficiency of client-side video conversion. You could probably do it in js these days or very worst case scenario have a cross platform desktop / phone client. WASM makes this an even easier task -- with some effort you could probably compile ffmpeg to WASM and run it fully client side in the browser with no performance penalty. If you take video conversion out of the equation, all that's left is file hosting, which is commoditized as possible these days, and the search/suggestions algorithm, which pisses people off with its accuracy anyway.
- wutbrodo 7y ago
  > Kubernetes, S3, Docker, and a whole slew of things that make this task fairly trivial compared to how it was in say 2005. Agreed, it's much _easier_ to do these things, but evidently not yet _easy_. Handwaving about who deserves credit is irrelevant compared to an analysis of whether executing on it is easy or not, and my (admittedly low-confidence) model,based on the empirics of the situation, is that its evidently still not that easy to reach the reliability and quality of YouTube as a service.

sam0x17 7y ago

Right but all of that auto scaling has been essentially donated to the community in the form of google cloud platform and AWS. Now we have cloud functions / lambdas (that are capable of running ffmpeg), Kubernetes, S3, Docker, and a whole slew of things that make this task fairly trivial compared to how it was in say 2005. There also isn't any rule saying you have to do your video conversions server side. Imagine the cost efficiency of client-side video conversion. You could probably do it in js these days or very worst case scenario have a cross platform desktop / phone client. WASM makes this an even easier task -- with some effort you could probably compile ffmpeg to WASM and run it fully client side in the browser with no performance penalty. If you take video conversion out of the equation, all that's left is file hosting, which is commoditized as possible these days, and the search/suggestions algorithm, which pisses people off with its accuracy anyway.

wutbrodo 7y ago
> Kubernetes, S3, Docker, and a whole slew of things that make this task fairly trivial compared to how it was in say 2005. Agreed, it's much _easier_ to do these things, but evidently not yet _easy_. Handwaving about who deserves credit is irrelevant compared to an analysis of whether executing on it is easy or not, and my (admittedly low-confidence) model,based on the empirics of the situation, is that its evidently still not that easy to reach the reliability and quality of YouTube as a service.