Are there any illustrations of the volumes of traffic that are being sent through core networks for netflix? I was under the impression the vast majority of traffic was served at the edge.
That's what they say now. Give them a few weeks and they'll start squealing. Every ISP oversubscribes for consumer segments. In the UK I guess not enough people are quarantined.
That has changed somewhat of late. With WBC there's no more "fixed" contention ratios and you pay for the last mile and then peer at national aggregation pops.
It's not true to say that A&A don't oversubscribe. Instead the situation is that they're happy to buy more capacity to fulfil their offer, within reason.
If A&A subscribers could in theory move 10Tbps (if they all simultaneously did some sort of download from a hypothetical unlimited source) but in reality they never do and it peaks at something like 100Gbps, A&A are fulfilling their promise by ensuring they've got 100Gbps to do that.
Typically you'll never notice the difference, except on your bill, because if they had 100x more bandwidth upstream they'd pay a lot of money for that, even though it was unused and they'd have to pass that to you in the medium term.
However, the reality is a little closer to oversubscribing. Suppose A&A are paying for 10Gbps on a particular port somewhere, and at their busiest time of the week it typically runs at 9.8Gbps. Unfortunately the company selling it only wants to offer 100Gbps as the next step up, for five times the money. Another 10Gbps port isn't an option as there are no 10Gbps ports free. So A&A decides to sit on the problem, nobody is suffering at 9.8Gbps.
Next week it hits 10Gbps and seems likely it'd have gone to 11Gbps if that was possible. Oops. The firm they're buying that capacity from still has 100Gbps available, but they agree at last that they could add more 10Gbps ports, and will do so for the same price as the existing port. Unfortunately it means buying a new Cisco router, which Cisc says is on back order, it'll arrive in July.
Are A&A going to throw five times the money at the problem for this burst of maybe 30-40 minutes per week? Or are they going to tell you sorry but it'll be July and until then bandwidth at that peak across that particular link isn't what it ought to be? They're going to do the latter. Because at the end of the day it's a business. RevK is a good guy, but he's not looking to bankrupt the company to make some point.
> In the UK I guess not enough people are quarantined.
A bit of that, and also probably a bit that their network needs and expectations, even when self-isolating, are lower than in the suburbs of tech-intensive Seattle.
Just note, in your service contract the measure of whether service is working or not will likely be limited to being able to reach your ISPs website. As long as you can do that, it’s just the vagaries of the internet, other providers you know.
It's perverse because it's harmful to both Netflix and the ISP's customers. The only reason ISPs get away with it is because they hold a monopoly on their user's internet access. If ISPs were subject to competition it would be a no brainer to host a Netflix cache within their network.
With my (relatively small) Cable ISP in an affected country it obvious to me that any Netflix content gets served from very close to the edge, with higher bandwidth and lower latency than almost any other content from the internet.
The last mile does not appear to be close to oversubscribed either, as indicated by my firewall which tracks RTT to the first hop (which is interestingly trending down compared to the past weeks) and the occasional speedtest that never drops below nominal bandwidth (200 Mbit).
If anything it is low latency livestreaming content (i.e. Youtube, Twitch, Mixer) that should be throttled, particularly over cellular networks.
I assume it's a matter of scale.
There aren’t, because this isn’t a problem with Netflix. It’s a problem with cheap ISPs that have gotten away with overselling service because most of their customers didn’t actually use it.
