These days, there are lively public internet exchanges up and down both coasts, in texas and chicago and elsewhere. A well placed backhoe can still make a big mess, many 'redundant fibers' are in the same strand, and last mile is fragile, but if my ISP network is up to the local internet exchange, there are many distinct routes to the other coast and a fiber cut is unlikely to route my traffic around the world.
I bet it's a lower number than anyone's actually comfortable with, though it would be rather difficult to pull of.
We've seen this one before... BGP doesn't (usually) look at link utilization, just link status and maybe that the BGP connection doesn't time out, but even a massively overloaded connection tends to have room for BGP pings, so it's less netsplit and more blackhole.
I don't think BGP hijacks is really the way to acheive this one.
edit: disk is for diskette
Call it BBP Protocol... "Blue Ball Prevention"
(1) https://atlantic.ctvnews.ca/mobile/many-atlantic-canadians-l...
"A few hubs"? Just Hurricane Electric has presence in many, many IXPs, and they're not even the largest transit provider:
* https://bgp.he.net/AS6939#_ix
AT&T, the largest transit provider, has a ridiculous number of peers:
* https://bgp.he.net/AS7018#_peers
And there are several different major global providers:
* https://en.wikipedia.org/wiki/Tier_1_network
A lot of the Big Tech companies are also building their own private fibre networks so they don't even have to worry about sharing infrastructure with the telcos:
Source: I was sitting next to the person who fixed it.
I had a long lost picture when they were doing some street construction outside and marked all the buried cables with spray paint. Lines going everywhere…
Next off, the their front page says:
> Carrier-Neutral with Access to 200+ Carriers
You're telling me these 200+ carriers have no other POPs?
Going to their "Connectivity" page, they show four trans-Pacific cable that land there:
* https://one-wilshire.com/connectivity/
Meanwhile there are a whole bunch of other trans-Pacific cables landing on the US West Coast:
* https://www.submarinecablemap.com
As someone who lives in Toronto, I certainly worry about how concentrated TORIX is, but even it has three physical locations (IIRC), and most folks connect elsewhere as well from the spelunking I've done in BGP.
So if one of these major carrier hotels does have a problem, I don't doubt that there would be repercussions, but I'm not worried about "the Internet" as a whole.
But look I live in Hong Kong, I dont feel that way: there are other backends, we could survive without the caching for a while and Google is forbidden for 1.4bn people who get on with it very well...
Depends what you call the internet. Yes, facebook and whatsapp are gone the minute one of those 3 companies screwed up.
Bing might be able to auto-scale using Azure infrastructure but there might not be enough hardware even there.
Yes, for the majority of users the two may be the same, but "The Internet" is the IPv4 and IPv6 network on which the web runs. The equivalent companies to what you named are the transit network operators, companies like Hurricane Electric, AT&T, CenturyLink, etc. The list of transit networks is very long, much longer than the list of major web hosting companies.
Now, that said, it is absolutely true that there are ways where major outages can be the result of just one failure. Last year there was a major outage on the US east cost after a Verizon fiber optic line was severed in NYC; it turns out that this line carried many supposedly redundant links, all bundled into a single cable. The failure of that line then caused a cascade of failures as rerouted traffic began overwhelming other systems, but in the end the outage was contained in a relatively small geographic region, albeit one with many Internet users.
Internet is much more reliable now.
https://en.m.wikipedia.org/wiki/GARR
I studied CS in Rome at "La Sapienza" and I was merely 10 steps away from INFN (national institute for nuclear physics) where GARR phisically resided.
I spent more time there than in my class.
That's how I got involved in this new "internet thing".
The "gigalapse" seemed like a big deal because proportionally a billion user hours in 1995 is everything doesn't work for days, or most things don't work for several weeks... but today of course it could also be everybody in the world is inconvenienced for about long enough to make breakfast. Oh no.
The larger tlds aren't quite as diversely hosted and certainly aren't amenable to long term caching, but it should take a major f up to break those too.
Some of the minor tlds, even the more popular ones do screw up from time to time though.
I dunno man, My used T420 laptop is serving several sites over a residential symmetrical fiber connection just fine.
Since about 1999 I have never NOT been running a server of some kind off my home connection. I wouldn't run a business off of it that way, but it's reliable enough to count on it which has to meet any sane definition of semi-reliable. The two biggest problems I've had have been essentially unrelated to the internet. The first is when I've been violating TOS of the ISP or the power was unreliable in the place I lived.
From yesterday. Note the bit about being on a Pi. Stood up quite admirably to an HN hug.
From 1998 to 2004 I served a fairly large amount of traffic out of my homebuilt pentium in a trailer out in the woods with an ISDN line. We stood up to several media mentions OK.
It was down when I tried it yesterday.
Download being 300Mpbs but upload 20Mbps IS kind of irritating though.
When did 100Mbps become popular for home LANs even?
I'm pretty sure new PCs and laptops have had gigabit standard for probably about 10 years.
Enthusiast and prosumer motherboards are now coming with 2.5 gigabit networking.
Also, smartphone proliferation made streaming from home NAS very convenient, as well as home security cameras and smart home features.
I disagree that current upload capacities are adequate. With 1Gbps+ upload connections standard, we might actually see privacy forward solutions that do not require us to depend on cloud services.
These systems could handle a ton of traffic very reliably. Consider that there was very little dynamic content and what there was barely taxed the CPU at all (e.g. a Perl CGI script to query and display some tiny amount of data from (LOL!) an Access database.