Total eclipse of the Internet: traffic impacts in Mexico, the US, and Canada | Hacker News Reader

Total eclipse of the Internet: traffic impacts in Mexico, the US, and Canada

blog.cloudflare.com

115 pointsjgrahamc63 comments

Reminds me of the first lockdown in the UK - I was on-call for a major food delivery service. We all went out at 7pm to clap for the NHS (which my NHS working family hated), and PagerDuty went off.

Went inside, apparently UK traffic had dropped by 20%. People from around the world were trying to work out why.

"Er, it's because we all put our phones down to clap for the nurses..."

Yes, that was super visible to us also: https://blog.cloudflare.com/when-people-pause

I always enjoy looking at Ramadan in the charts also: https://blog.cloudflare.com/how-ramadan-shows-up-in-internet...

Nice. Ramadan / Eid is difficult as it varies so you can't put up an accurate notice earlier for on-call engineers.

It's based on the Hijri calendar (based on the moon) and starts 11 days earlier each year in Gregorian calendar.

But the time changes depending on where you live

Aerbil, are you from from Erbil by any chance?

No. Please use email for personal questions.

> which my NHS working family hated

Why?

Probably because it's the very definition of virtue signaling instead of actually supporting them in meaningful ways, e.g. a pay raise

I was working for the NHS at the time. Peak lockdown for me was doing my weekly shop in Asda during the national weekly clap...

I had to queue to get in to the store because at the time social distancing rules meant only so many people were allowed to shop at any given time. Once the Covid marshal granted me entry I washed my hands as required with anti bacteria hand wash and put on my disposable gloves. I then proceeded to carefully navigate the store ensuring I avoided sharing an aisle with any other shoppers to avoid attracting the attention of one of the many Covid marshals positioned around the store ensuring shoppers were adhering to Covid policy. While shopping the tannoy system repeatedly reminded me that I must remain masked and to keep social distance from other shoppers. I walked over to the Pizza aisle to grab a Pizza, but most were sold out. A few remained though, however they were all NHS themed pizzas to support "our NHS". Then the tannoy system announced to customers that the weekly scheduled national NHS clap was about to begin and that shoppers were encourage to take part. I completely forgot about this... I'm holding a basket in one hand and 2L bottle of Pepsi in the other. Do I put down my items and wait for the national clap to begin or should I just continue shopping as if it's not happening? No, what am I thinking, not clapping is not an option. I put down my items and wait for the clapping to begin. Then we all clapped before proceeding with our shopping.

Good times.

This "clapping" thing does sound a bit 1984-like, a ritual to remember that others are watching you and so you should conform to the social norm.

Did the clapping actually help the NHS in any way?

It was intended to send a message of “we support you”.

However, most healthcare workers are paid incredibly poorly and the NHS is generally in an extremely poor state, in part due to mismanagement and underfunding by successive governments. Many healthcare workers feel their lives could be made much better at the voting booth, rather than by people clapping at their door and continuing to vote for a party that has a history of (justly or not) battling against even below-inflation pay-rises. As ever, other factors are at play and the issue is almost certainly not as simple as blaming the government. Nonetheless, there is a general feeling that the Conservative-led government policies since 2010 have reduced the standard of living for many, especially healthcare workers.

The UK is poised to have a change in governing party for the first time in 14 years, at least, but the clapping was seen by many at the time as hypocritical of the voting public.

In Turkish there is the phrase "göz boyama" translating to something like "window dressing" or similar. The UK government gave contracts worth billions to a scam company for pandemic related 'Track & Trace' work, which all proved to be shenanigans, but when it came to actually rewarding the hard-working NHS heroes they got applauds and painted papers on people's windows. They had been asking for a fair pay rise even before the pandemic.

The UK government has brought one of the best free national health systems in the world to the point of being worse than useless in a span of about a decade.

which my NHS working family hated

I am not surprised at all.

Oh, neat! I worked for a Car Load Financing company during the pandemic. We had data on how many people were applying for loans by geography, and how much it dipped from the "normal" load. I didn't realize it at the time, but areas with a "low" dip also repealed their stay-at-home orders sooner.

Unpopular opinion: I miss the lockdowns. It was empty, clean, no traffic, no delays, excuse is ready for introverts, list goes on!

It's not an unpopular opinion, a lot of people like things to be quiet and peaceful; if it wasn't for job opportunities and boredom I suspect a lot of people would prefer to live in the countryside (in the broadest sense of the word). Only a few people thrive in busy circumstances, those were the ones that were climbing up the walls after two weeks of lockdown.

Spot on! I’m glad to see some people agreeing with that. Around me, whenever lockdowns are mentioned, they are recalled as dark times.

Why can't tech companies seem to keep this stuff to themselves? I mean, it's interesting and all, but it's an even bigger reminder that Cloudflare is an Mitm attack on the Internet, in order to have access to this data. And if they have that data, no one could possibly believe they aren't somehow monetizing it. Consider if we read the same post from Google or Facebook; it would not be as well recieved. When the Internet was still little, it was cute to read OkCupid's data blog but those days are long past, especially after leaks about Uber's heaven view. The post by Sleep8 in the wake of the OpenAI drama was also poorly recieved.

Corporate blog posts are advertising for their brand and service. If they happen to overlap with something interesting that's a bonus.

I'm wondering if this might have been a result of wireless internet traffic being overwhelmed with everyone in the areas of totality uploading pictures and video... actually using internet bandwidth at the exact same moment? The resulting congestion reducing overall internet usage because it was focused on wireless networks only vs/ broadband home connections (since obviously everyone was outside and not inside)

It's possible, but we don't have a state-level perspective on mobile devices vs desktop, only at the country level. So, it's difficult to confirm.

doubtful. people were busy watching it, not live streaming it.

Internationally people would be streaming it, but TBH the images of the thing itself aren't that impressive and the video stream compresses well. I did like seeing a timelapse of a city as the eclipse happened.

Related: a similar analysis based on the 2017 eclipse. It has a map with zip-code level granularity and a comparison of desktop vs mobile behavior:

https://web.archive.org/web/20171003232834/https://www.comsc...

Note that this decrease in traffic is after a huge influx of visitors. Highways in and out of the area of totality were overwhelmed in most areas. I live in Michigan and drove 2 hours south to find myself in a small town park in Ohio surrounded by other Michigan residents.

Heh, I could've written this comment. I also found myself 2 hours south in a small Ohio park.

It would be interesting to see a map of traffic by state, or even better by county, compared to normal traffic the morning of the eclipse and the evening before the eclipse.

A heck of a lot of people travelled to the eclipse and so their normal Sunday evening/Monday morning internet use should be missing from where they came from. At their destination they will be using internet to find things like food or to pass the time waiting for the eclipse so should be adding to the normal Sunday evening/Monday morning internet traffic there.

Reminds me of water consumption spiking during the football world cup interval (toilet flushes).

Fascinating info, and great visualization!

An interesting discontinuity that jumps out to me is that AR dropped 54%, but neighboring LA and MS dropped 0%. AR was in totality, but LA and MS were close.

Part of it is probably related to the distribution of population vs the path of totality. Similarly, VT dropped more than double what neighbor NY did, but VT's population is mostly in Burlington (totality) whereas NY's is in NYC (not totality and not so mobile).

But 54% to 0% seems too dramatic a difference for that to be the full explanation. Any insights? Maybe just inadequate data to calculate?

Totality is a much bigger deal.

Near totality is cool, but it's not the same couple of minutes of magic.

Agreed, but the other neighboring, non-totality, states, do not show the same discontinuity in data.

Hey there. I can provide some more details. Our data doesn't show as much of a drop in Mississippi compared to Arkansas. However, in cases where traffic is higher than usual during the day (as it was in Mississippi and Louisiana), the drop compared with the previous week is even less noticeable using this method.

Interesting, I was wondering about that too -- so are the "0% drop" states sometimes actually "non-zero % increase" states?

I think there’s only one or two states where that happens. If you check the table, I use “—“ instead of 0%.

Louisiana was pretty smothered by clouds during eclipse time yesterday which likely had an impact.

In my country, the largest internet provider had intermitent service interruptions for a couple hours ADSL and mobile. They were approximately 1 minute every two minutes. Even 3g failed a couple times.

Such interruptions happen a couple times a year, only. Coincidence % would be 2(100)/(365)

We had a 20% eclipse so, it was a regular working day. I would rule out increased usage)

"Close" on a cloudy day (which it was for most of LA and MS)? Not a lot to see, especially when you're all still at work and there's a live stream from NASA of a variety of locations - and you're watching it.

Whereas a huge proportion of Arkansas' population was in totality (and presumably watching outside). About the only larger place that didn't experience it was the Fayetteville-Bentonville corridor.

Akamai should also have visibility into the global traffic patterns. It would be nice to know if their data matches with cloudflare.

Not just Akamai, even Netflix and other streaming services might have noticed this drop.

In Brazil, during the World Cup, it's very common for employers to give the day off when Brazil is playing, or at the very least let people take a break to watch the game in the office.

In 2010, I was working for a network management company, and we kept one eye on the networks, another on the game. All networks graphs saw a very large impact when the game was on, with a quick surge during half time. It was fucking unreal. I believe this wouldn't happen anymore nowadays, given the pervasiveness of streaming services etc.

Anecdotally, T-Mobile cellular internet slowed to a crawl in southern Illinois when the eclipse entered the partial stage and was flakey for about an hour afterwards.

AT&T in SE Oklahoma was practically unusable as well. Ping times were 300-1000ms and packet loss was (didn't run ping for very long) probably 90+%. Loading a webpage just didn't work. Sending a WhatsApp text message worked if you were patient.

It's interesting that it was so closely tied to the actual eclipse. I'd have not expected that. I'd have expected people to be trying to make the most use of cellular while waiting for the eclipse. You've got all those people who travelled to it trying to keep themselves occupied while waiting.

I saw the 2017 eclipse near Madras, OR. That part of Oregon is mostly agricultural with scattered small towns every few miles. T-Mobile was close to useless from when I got there around noon the day before the eclipse to at least an hour after the eclipse.

An old friend that I hadn't seen in decades was in the same general area and we both had T-Mobile. We were trying to figure out if we could meet up, and I don't think we managed to get a single voice call to go through. SMS would occasionally go through albeit with an hour or more of lag.

Many more people taking photos and videos than usual, and sharing those photos and videos, as well as iCloud automated backup?

I assume friends/family calling/Facetiming to see or share the experience as well. Many traveled there and didn't have WiFi.

iCloud Backup generally happens at night when on Wi-Fi.

I really want to see an analysis of the impact on the power grid. A friend shared their solar panel numbers and it was pretty cool

I wonder just how many images of the eclipse were taken and then uploaded as a result...

I witnessed one years ago and was completely awe-struck by it - or more, how tiny we all are in the grand scheme of things.

And we sit here on our chairs getting excited about data logs!

How does Cloudflare know what the searches were made over the internet ? I am assuming all communication happening over search engines is TLS protected.

There's nothing in this blog about searches. It's just raw traffic data across all the sites/services that use us. So this is showing overall change in traffic while the eclipse was happening and isn't related to what people were actually doing on the Internet.

What do you mean by searches? The article talks about internet traffic in general, which they can monitor.

Cloudflare MITM lots of connections plus they run a DNS server so they know exactly what you're up to.

Running a DNS server lets them know what site you're going to, just like everyone who sees the secured request also knows, right? Otherwise there'd be no way for the request to end up at the right location. I suppose they might have an advantage with an encrypted DNS, but I don't think it's much of a MITM if you're sending your request to them to find out where to go.

Cloudflare terminates SSL for all customers behind their DDOS protection, by necessity.

GP is technically correct, but is tilting at windmills at best, making veiled accusations at worst.

Oh, interesting. I didn't realize that.

Was it veiled? Sorry. Cloudflare is an arm of the US government helping to eavesdrop on all internet traffic.

All US companies follow US law. It sounds goofy, I know, but it's a cultural thing.

When you operate at Cloudflare's scale, you receive a lot of inquiries from law enforcement in all countries. See also Google, Microsoft, Amazon, Apple, AT&T, Verizon, Linode, Hurricane Electric, et cetera ad inifinitum.

If your accusation is more specific than that, you'll want to rethink the word "arm" which is clearly incorrect and makes you sound like a loon.

New account "just asking questions".

Thanks Boss, totally wasn’t the intention. But I mistakenly assumed traffic was all related to “search”.

Now I am not a fool anymore by “just asking question”.

Really cool parallax effect on that top image when I swipe back. How is that done?

Yea that 10 hour drive from upstate NY was brutal! The traffic was a nightm....opens article

Oh yea and that.

Grammatically the title should be "Total eclipse ON the Internet:"

This title is obviously a play on the Bonnie Tyler song "Total Eclipse of the Heart".

Because the Internet is Love.

Exactly. That was the inspiration :)

Reply on news.ycombinator.com