> Why it is hard for Netflix. They have already figured out CDN part. So it should not be a problem even if it is 1M or 100M. because their CDN infrastructure is already handling the load ... Note: I am not saying my 100k is same scaling as their 100M. I am curious about which part is the bottleneck.
100k concurrents is a completely different game compared to 10 million or 100 million. 100k concurrents might translate to 200Gbps globally for 1080p, whereas for that same quality, you might be talking 20T for 10 million streams. 100k concurrents is also a size such that you could theoretically handle it on a small single-digit number of servers, if not for latency.
> CloudFront also has a feature request collapsing that greatly help live streaming.
I don't know how much request coalescing Netflix does in practice (or how good their implementation is). They haven't needed it historically, since for SVOD, they could rely on cache preplacement off-peak. But for live, you essentially need a pull-through cache for the sake of origin offload. If you're not careful, your origin can be quickly overwhelmed. Or your backbone if you've historically relied too heavily on your caches' effectiveness, or likewise your peering for that same reason.
200Gbps is a small enough volume that you don't really need to provision for that explicitly; 20Tbps or 200Tbps may need months if not years of lead time to land the physical hardware augments, sign additional contracts for space and power, work with partners, etc.
100k concurrents is a completely different game compared to 10 million or 100 million. 100k concurrents might translate to 200Gbps globally for 1080p, whereas for that same quality, you might be talking 20T for 10 million streams. 100k concurrents is also a size such that you could theoretically handle it on a small single-digit number of servers, if not for latency.
> CloudFront also has a feature request collapsing that greatly help live streaming.
I don't know how much request coalescing Netflix does in practice (or how good their implementation is). They haven't needed it historically, since for SVOD, they could rely on cache preplacement off-peak. But for live, you essentially need a pull-through cache for the sake of origin offload. If you're not careful, your origin can be quickly overwhelmed. Or your backbone if you've historically relied too heavily on your caches' effectiveness, or likewise your peering for that same reason.
200Gbps is a small enough volume that you don't really need to provision for that explicitly; 20Tbps or 200Tbps may need months if not years of lead time to land the physical hardware augments, sign additional contracts for space and power, work with partners, etc.