8 ms·

My beef was not more like "who did it first", I honestly don't know and I couldn't find a reference online of anyone did something like this at this scale befor

by pavs 9y ago

My beef was not more like "who did it first", I honestly don't know and I couldn't find a reference online of anyone did something like this at this scale before google/youtube/akamai. The Point is the global scale of this on-premise network and the amount of data they move. I think Netflix is a blip compare to youtube/google/Akamai/facebook (I recently learned facebook also has a similar program, but I think they also use Akamai), on a global scale. I didn't like that glossed on over this as if Netflix is doing something exceptional or even unique. Obviously, it takes some serious engineering chops scale this amount of traffic globally and routes the traffic through on-premise nodes. But at this point, it's not exceptional - if anything its commodity.

Terretta 9y ago
Netflix isn't a blip. For instance: https://fortunedotcom.files.wordpress.com/2015/10/apps-by-internet-traffic.png https://fortunedotcom.files.wordpress.com/2015/10/apps-by-in... Still true in 2017: https://techcrunch.com/2017/07/05/u-s-cord-cutters-watch-more-netflix-than-amazon-video-hulu-and-youtube-combined/ https://techcrunch.com/2017/07/05/u-s-cord-cutters-watch-mor... Regardless, long form video streamed globally without buffering is incredibly difficult. Short form clips are drastically easier. Do the math on how much content at what quality can fit in a given cache, or how long it takes to swap a piece of content in a cache, you'll start to see why. It's surprising to me that streaming well is still not a commodity. Even less so if you consider transactional (paid) video, which is what Netflix does. For example, just recently, tech press talked about cable companies losing $10B a year from account sharing. This was a problem VDNs solved in late 90s / early 00s for streaming protocols, but the technique for handling this across a global footprint was sidelined when YouTube's "http" progressive downloads / pseudo streaming and 'ad supported' video became vogue. Now that streams need to act more like streams again, and people want to charge for the content itself again, the solution is again needed, and no http-based "commodity" CDN seems to have it sorted out. Netflix does. PS. A couple refs from before Netflix/Google/YouTube: - https://www.google.com/patents/US20010027491 https://www.google.com/patents/US20010027491 - https://www.google.com/patents/US20010047275 https://www.google.com/patents/US20010047275 - https://news.microsoft.com/2000/06/12/microsofts-new-digital-broadcast-manager-enables-next-wave-of-e-commerce/ https://news.microsoft.com/2000/06/12/microsofts-new-digital... Those were first iterations. Later iterations were more sophisticated, and not patented in order to not disclose.
- pavs 9y ago
  Thanks for the references. I was referring to who did a large _volume of traffic_ in this type of distributed node first on a _global scale_ - not who did it / patented it first. Most importantly, outside the North-American/Europe echo chamber Netflix is a blip (a large blip nonetheless) compare to google/facebook/Akamai (probably even Cloudflare) traffic. I don't say this to undermine what Netflix does both as a service and a Network Engineering standpoint - I am personally a big fan of their service and their tech blog - I was trying to make a point on some info mentioned in the blog. I am sorry if I wasn't able to express my point properly.
  - Terretta 9y ago
    > I was referring to who did a large _volume of traffic_ in this type of distributed node first on a _global scale_ - not who did it / patented it first. I am too. For example, even back then we did live streaming events of millions of simultaneous viewers and billions of on-demand streams a month, from petabytes of content in 14 datacenters globally (including Southeast Asia). All before Netflix, Google, Facebook. Akamai and most of the top CDNs were clients. > Outside NA/EU echo chamber Netflix is a blip... Netflix is very much playing outside ‘the echo chamber’: http://www.popularmechanics.com/technology/a20138/netflix-interview-wired-traffic-exceeds-internet-capacity/ http://www.popularmechanics.com/technology/a20138/netflix-in... And growing: http://variety.com/2017/digital/news/netflix-250-million-hours-1202010393/ http://variety.com/2017/digital/news/netflix-250-million-hou... If you have stats that show they are ‘a blip’ in traffic volume globally, please share. > I was trying to make a point on some info in the blog By continuing to insist Netflix is “a blip”, and offering no definition of what you mean by large scale, or why you’re so sure Google is doing video delivery networking better, I’m finding it difficult to have a substantive discussion. PS. You started an ISP in Southeast Asia, and looks like you had Google/Akamai footprint for your ISP. So maybe your ‘blip’ assertion is your ISP data? From the provider point of view, different ISPs (even in US) had very different viewing patterns. Without looking into individual ISPs, we guessed this related to median connection speeds, peering headroom, and localized content/viewing preferences.

by pavs 9y ago

Terretta 9y ago

Netflix isn't a blip. For instance: https://fortunedotcom.files.wordpress.com/2015/10/apps-by-internet-traffic.png https://fortunedotcom.files.wordpress.com/2015/10/apps-by-in... Still true in 2017: https://techcrunch.com/2017/07/05/u-s-cord-cutters-watch-more-netflix-than-amazon-video-hulu-and-youtube-combined/ https://techcrunch.com/2017/07/05/u-s-cord-cutters-watch-mor... Regardless, long form video streamed globally without buffering is incredibly difficult. Short form clips are drastically easier. Do the math on how much content at what quality can fit in a given cache, or how long it takes to swap a piece of content in a cache, you'll start to see why. It's surprising to me that streaming well is still not a commodity. Even less so if you consider transactional (paid) video, which is what Netflix does. For example, just recently, tech press talked about cable companies losing $10B a year from account sharing. This was a problem VDNs solved in late 90s / early 00s for streaming protocols, but the technique for handling this across a global footprint was sidelined when YouTube's "http" progressive downloads / pseudo streaming and 'ad supported' video became vogue. Now that streams need to act more like streams again, and people want to charge for the content itself again, the solution is again needed, and no http-based "commodity" CDN seems to have it sorted out. Netflix does. PS. A couple refs from before Netflix/Google/YouTube: - https://www.google.com/patents/US20010027491 https://www.google.com/patents/US20010027491 - https://www.google.com/patents/US20010047275 https://www.google.com/patents/US20010047275 - https://news.microsoft.com/2000/06/12/microsofts-new-digital-broadcast-manager-enables-next-wave-of-e-commerce/ https://news.microsoft.com/2000/06/12/microsofts-new-digital... Those were first iterations. Later iterations were more sophisticated, and not patented in order to not disclose.

pavs 9y ago
Thanks for the references. I was referring to who did a large _volume of traffic_ in this type of distributed node first on a _global scale_ - not who did it / patented it first. Most importantly, outside the North-American/Europe echo chamber Netflix is a blip (a large blip nonetheless) compare to google/facebook/Akamai (probably even Cloudflare) traffic. I don't say this to undermine what Netflix does both as a service and a Network Engineering standpoint - I am personally a big fan of their service and their tech blog - I was trying to make a point on some info mentioned in the blog. I am sorry if I wasn't able to express my point properly.
- Terretta 9y ago
  > I was referring to who did a large _volume of traffic_ in this type of distributed node first on a _global scale_ - not who did it / patented it first. I am too. For example, even back then we did live streaming events of millions of simultaneous viewers and billions of on-demand streams a month, from petabytes of content in 14 datacenters globally (including Southeast Asia). All before Netflix, Google, Facebook. Akamai and most of the top CDNs were clients. > Outside NA/EU echo chamber Netflix is a blip... Netflix is very much playing outside ‘the echo chamber’: http://www.popularmechanics.com/technology/a20138/netflix-interview-wired-traffic-exceeds-internet-capacity/ http://www.popularmechanics.com/technology/a20138/netflix-in... And growing: http://variety.com/2017/digital/news/netflix-250-million-hours-1202010393/ http://variety.com/2017/digital/news/netflix-250-million-hou... If you have stats that show they are ‘a blip’ in traffic volume globally, please share. > I was trying to make a point on some info in the blog By continuing to insist Netflix is “a blip”, and offering no definition of what you mean by large scale, or why you’re so sure Google is doing video delivery networking better, I’m finding it difficult to have a substantive discussion. PS. You started an ISP in Southeast Asia, and looks like you had Google/Akamai footprint for your ISP. So maybe your ‘blip’ assertion is your ISP data? From the provider point of view, different ISPs (even in US) had very different viewing patterns. Without looking into individual ISPs, we guessed this related to median connection speeds, peering headroom, and localized content/viewing preferences.