How did we get to a place where either Cloudflare or AWS having an outage means a large part of the web going down? This centralization is very worrying.
Oddly this centralization allows a complete deferral of blame without you even doing anything: if you’re down, that’s bad. But if you’re down, Spotify is down, social media is down… then “the internet is broken” and you don’t look so bad.
It also reduces your incentive to change, if “the internet is down” people will put down their device and do something else. Even if your web site is up they’ll assume it isn’t.
I’m not saying this is a good thing but I’m simply being realistic about why we ended up where we are.
The internet can’t afford to just “give people mental health breaks.”
What happened to having a business continuity plan? E.g. when your IT system is down, writing down incoming orders manually and filling them into the system when it's restored?
I have a creeping suspicion that people don't care about that, in which case they can't really expect more than to occasionally be forced into some downtime by factors outside of their control.
Either it's important enough to have contingencies in place, or it's not. Downtime will happen either way, no matter how brilliant the engineers working at these large orgs are. It's just that with so much centralization (probably too much) the blast range of any one outage will be really large.
Doesn't change the fact that 99% of our ticket sales happen online. People will even come in to the theatre to check us out (we're magicians and it's a small magic shop + magic-themed theatre - so people are curious and we get a lot of foot traffic) but, despite being in the store, despite being able to buy tickets right then and there and despite the fact that it would cost less to do so ... they invariably take a flyer and scan the QR code and buy online.
We might be kind of niche, since events usually sell to groups of people and it's rare that someone decides to attend an event by themselves right there on the spot. So that undoubtedly explains why people behave like this - they're texting friends and trying to see who is interested in going. But I'm still bringing us up as an example to illustrate just how "online" people are these days. Being online allows you to take a step back, read the reviews, price shop, order later and have things delivered to your house once you've decided to commit to purchasing. That's just normal these days for so many businesses and their customers.
If AWS is down, most businesses on AWS are also down, and it’s mostly fine for those businesses.
It's better to have diverse, imperfect infrastructure, than one form of infra that goes down with devastating results.
I'm being semi-flippant but people do need to cope with an internet that is less than 100% reliable. As the youth like to say, you need to touch grass
Being less flippant: an economy that is completely reliant on the internet is one vulnerable to cyberattacks, malware, catastrophic hardware loss
It also protects us from the malfeasance or incompetence of actors like Google (who are great stewards of internet infrastructure... until it's no longer in their interests)
The idea that we absolutely need 24/7 productivity is a new one and I’m not that convinced by it. Obviously there are some scenarios that need constant connectivity but those are more about safety (we don’t want the traffic lights to stop working everywhere) than profit.
We don't need it, the owners want it
Phone lines absolutely did not go down. Physical POTS lines (Yes, even the cheap residential ones) were required to have around 5 9s of availability, or approximately 5 minutes per year. And that's for a physical medium affected by weather, natural disasters, accidents, and physical maintenance. If we or the LEC did not meet those targets contracts would be breached and worst case the government would get involved.
Physical network equipment is redundant and reliant enough that getting 5 minutes of downtime or less per year is totally doable.
the web however... is a far different beast (and in my opinion, with an incentive which does not factor in reliability)
This was the same when I was doing consulting inside (ie large companies willing to pay the premium cost of AWS ProServe consultants) and outside working at 3rd party companies.
try going outside
This:
> if you’re down, that’s bad. But if you’re down, Spotify is down, social media is down… then “the internet is broken” and you don’t look so bad.
is just marketing. If you are down with some other websites it is still bad.
In some cases, absolutely. For the vast majority, it really, really doesn't matter.
(Source: my personal website is down and nobody cares, including me)
> it is still bad
No doubt. But there’s a calculation to make, is it bad enough to spend the extra money on mitigations, to hire extra devops folks to manage it all… and in the majority of end user facing cases the answer is no, it isn’t.
That one time when an AZ goes down and your infra successfully fails over to the other two isn't worth it for a lot of my scale companies, ops consultants seem to be chasing high cloud spend to justify their own high cost. I also factor in that I live in Sweden where most infrastructure outages are exceptionally rare.
Ofc it depends on what kind of company you are and what you're providing.
In this case, the internet should be down more often.
Users are never a consideration today anyway.
We had very decentralized "internet" with BBSes, AOL, Prodigy, etc.
Then we centralized on AOL (ask anyone over 40 if they remember "AOL Keyword: ACME" plastered all over roadside billboards).
Then we revolted and decentralized across MySpace, Digg, Facebook, Reddit, etc.
Then we centralized on Facebook.
We are in the midst of a second decentralization...
...from an information consumer's perspective. From an internet infrastructure perspective, the trend has been consistently toward more decentralization. Initially, even after everyone moved away from AOL as their sole information source online, they were still accessing all the other sites over their AOL dial-up connection. Eventually, competitors arrived and, since AOL no longer had a monopoly on content, they lost their grip on the infrastructure monopoly.
Later, moving up the stack, the re-centralization around Facebook (and Google) allowed those sources to centralize power in identity management. Today, though, people increasingly only authenticate to Facebook or Google in order to authenticate to some 3rd party site. Eventually, competitors for auth will arrive (or already have ahem passkeys coughcough) and, as no one goes to Facebook anymore anyway, they'll lose grip on identity management.
It's an ebb and flow, but the fundamental capability for decentralization has existed in the technology behind the internet from the beginning. Adoption and acclimatization, however, is a much slower process.
That's the whole issue with monopolies for example, innit? We envision "ideal free market dynamics" yet in practice everybody just centralizes for efficiency gains.
The much bigger issue with monopolies is that there is no pressure on the monopolist to compete on price or quality of the offering.
I don't have a better solution, but it's a clear problem. Also, for some reason, more and more people (not you) will praise and attack anyone who doesn't defend state A (ideal equilibrium). Leaving no room to point out state B as a logical consequence of A which requires intervention.
If anyone in the industry actually cared about reliability and took personal stake in their system being up, everyone would be back on-prem.
No, the value proposition was always about saving money, turning CapEx into OpEx. Direct quote from my former CEO maybe 9 years ago: We are getting out of the business of buying servers.
Cloud engineering involves architecting for unexpected events: retry patterns, availability zones, multi-region fail over, that sort of thing.
Now - does it all add up to cost savings? I could not tell you. I have seen some case studies, but I also have been around long enough to take those with a big grain of salt.
IMHO it adds, but only if you are big enough. Netflix level. At that level, you go and dine with Bezos and negotiate a massive discount. For anyone else, I’d genuinely love to see the numbers that prove otherwise.
> There's zero personal responsibility
Unfortunately, this seems to be the unspoken mantra of modern IT management. Nobody wants to be directly accountable for anything, yet everyone wants to have their fingerprints on everything. A paradox of collaboration without ownership.
And this is not reserved instances, this is an org level pricing deal. Some have been calling it anti-competitive and saying the regulators need to look at the practice.
It adds if you're smart about using resources efficiently, at any level. And engineer the system to spin up / spin down as customers dictate.
For situations where resources are allocated but are only being utilized a low percentage (even < 50% in some cases), it is not cost effective. All that compute / RAM / disk / network etc. is just sitting there wasted.
You no longer needed them to approve a new machine, you just spun it up how you want. Sped things up massively for a while.
I saw this stuff too many times, and it is precisely why the cloud exploded in use in about 2010.
For many organizations, that's literally illegal, and anyone who does this should be fired.
But sure jump to more conclusions if you want.
And this is in Fortune 500 of course.
I mean yea, but who knows how long that box would sit around before it was discovered.
If the business can live with a couple of hours downtime per year when "cloud" is down, and they think they can ship faster / have less crew / (insert perceived benefit), then I don't know why that is a problem.
Frankly it's a blessing, always being able to blame the cloud that management forced company to migrate to be "cheaper" (which half of the time turns out to be false anyway)
https://www.apple.com/newsroom/2025/11/apple-introduces-digi...
But we can all say thank you to all the AI crawlers who hammer websites with impossible traffic.
I.e., if you use Tor for "normie sites", then the fact that someone can be seen using Tor is no longer a reliable proxy for detecting them trying to see/do something confidential and it becomes harder to identify & target journalists, etc. just because they're using Tor.
If you have a site with valuable content the LLM crawlers hound you to no end. CF is basically a protection racket at this point for many sites. It doesnt even stop the more determined ones but it keeps some away.
And they're pretty tame as far as computer fraud goes - if my device gets compromised I'd much rather deal with it being used for fake YouTube views than ransomware or a banking trojan.
"The internet sucks", yes, but we're doing it to ourselves.
LLMs aren't going anywhere, but the world would be a better place if they hadn't been developed. Even if they had more positive impacts, those would not outweigh the massive environmental degradation they are causing or the massive disincentive they created against researching other, more useful forms of AI.
LLMs have become a crucial compendium of knowledge, that had become hidden behind SEO
A solid secondary option is making LLM scraping for training opt-in, and/or compensating sites that were/are scraped for training data. Hell, maybe then you could not knock websites over incentivizing them to use Cloudflare in the first place.
But that means LLM researchers have to respect other people's IP which hasn't been high on their todo lists as yet.
bUt ThAT dOeSn'T sCaLe - not my fuckin problem chief. If you as an LLM developer are finding your IP banned or you as a web user are sick of doing "prove you're human" challenges, it isn't the website's fault. They're trying to control costs being arbitrarily put onto them by a disinterested 3rd party who feels entitled to their content, which it costs them money to deliver. Blame the asshole scraping sites left and right.
Edit: and you wouldn't even need to go THAT far. I scrape a whole bunch of sites for some tools I built and a homemade news aggregator. My IP has never been flagged because I keep the number of requests down wherever possible, and rate-limit them so it's more in line with human like browsing. Like so much of this could be solved with basic fucking courtesy.
Most of the problems on the internet in 2025 aren't because of one particular technology. They're because the modern web was based on gentleman's agreements and handshakes, and since those things have now gotten in the way of exponential profit increases on behalf of a few Stanford dropouts, they're being ignored writ large.
CF being down wouldn't be nearly as big of a deal if their service wasn't one of the main ways to protect against LLM crawlers that blatantly ignore robots.txt and other long-established means to control automated extraction of web content. But, well, it is one of the main ways.
Would it be one of the main ways to protect against LLM web scraping if we investigated one of the LLM startups for what is arguably a violation of the Computer Fraud and Abuse Act, arrested their C-suite, and sent each member to a medium-security federal prison (I don't know, maybe Leavenworth?) for multiple years after a fair trial?
Probably not.
It is the Web that is being degraded
The site doesn't even need to have valuable content. Any content at all.
That's a problem caused by bots and spammers and DDoSers, that Cloudflare is trying to alleviate.
And you generally don't have to prove it over and over again unless there's a high-risk signal associated with you, like you're using a VPN or have cookies disabled, etc. Which are great for protecting your privacy, but then obviously privacy means you do have to keep demonstrating you're not a bot.
That they're trying counts for brownie points, it's not an excuse to be satisfied with something that still bothers a lot of people. Do better, CloudFlare.
If you have any ideas on how to protect against bad actors in a way that is just as effective but easier for users, please share it.
Because as far as I can tell, this isn't a question of effort. It's a question of fundamental technological limitations.
If you have a better technological solution, we'd all love to know it. Because right now, site owners are using the best tools available.
Criticizing when there's no other solution isn't very useful, is it?
Cloudflare is the multi-billion dollar corporation. It has everything to do with that, because they are the primary cause, and their resources and position make them by far the best equipped to solve it.
> Criticizing when there's no other solution isn't very useful, is it?
Of course it is. Without criticism, the growing problem goes unacknowledged and allowed to persist. It should instead be continually called out until it is prioritized, and some of those billions should be spent on researching a solution. (Similarly, a company found to be dumping waste into a river should be held responsible for cleaning up the mess they created. Even if that turns out to be expensive or difficult.)
Expecting a single affected person to solve it for the big corp that caused it is unrealistic. And blaming the victims because they use VPNs or disable cookies is... unhelpful.
CloudFlare is protecting sites from DDoS attacks and out-of-control bots. They're not the ones causing them. If CloudFlare wasn't asking you to prove you're human, many times the site would be down entirely because it couldn't keep up. Or the site would simply shut down because it couldn't afford it.
And this isn't a question of spending some fraction of billions on researching a solution. There fundamentally isn't one, if you understand how the internet works. This is a problem a lot of people would like to solve better, believe me.
So, yes, criticizing Cloudflare here is as useful as criticizing it for not having faster-than-light communication. There's nothing else it can do. It's not "blaming the victims".
I'm going to assume you simply don't have the technical understanding of how the internet works. Because the position you're taking is simply absurd and nonsensical, and there's no way you would write what you're writing otherwise.
As long as HN is up and running, everything is going to be O.K.!
Smaller companies that provide real world services or goods to make a much more meagre living that rely on some of the services sold to them by said software companies will be impacted much more greatly.
Losing a day or two of sales to someone who relies on making sales every day can be a growing hardship.
This doesn’t just impact developers. It’s exactly this kind of myopic thinking that leads to scenarios like mass outages.
You have to realize when software companies tell the world they should rely on their works, they world will do so. And once that occurs, the responsibility is all on the software companies to meet the expectations they built in people!
It's mad that this industry works so hard to claim the trust of millions of people, then shirks it as soon as it's convenient.
It's shameful.
When have users been asked about anything?
this is like a bad motivational speaker talk.. heavy exhortations with a dramatic lack of actual reasoning.
Systems are difficult, people. It is "incentives" of parties and lockin by tech design and vendors, not lack of individual effort.
The internet this day is fucking dangerous and murderous as hell. We need Cloudflare just to keep services up due to the deluge of AI data scrapers and other garbage.
In my direct experience, this isn't true if you're running something even vaguely mission-critical for your customers. Your customer's workers just know that they can't do their job for the day, and your customer's management just knows that the solution they shepherded through their organization is failing.
If akamai went down i have a feeling you'd see a whole lot more real life chaos.
if you run anything even remotely mission critical, not having a plan B which is executable and of which you are in control (and a plan C) will make you look completely incompetent.
There are very, very few events which some people who run mission critical systems accept as force majeur. Most of those are of the scale "national emergency" or worse.
And why should anyone be surprised? It's been about 80 years since "The buck stops here."[0] had any real relevance. And more's the pity.
[0] https://www.phrases.org.uk/meanings/the-buck-stops-here.html
End product users have no power, they can complain to support and maybe get a free month of service, but the 0.1% of customers that do that aren't going to turn the tide and have anything change.
Engineering teams using these services also get "covered" by them - they can finger point and say "everyone else was down too."
oh no
I agree. When people talk about the enshittification of the internet, Cloudflare plays a significant role.
Which changes nothing to you actually being down, youre only down more. CF proxies always sucked - not your domain, not your domain...
So there are massive economies of scale. Small CDN with (say) 10,000 customers and 10mbit/sec per customer can handle 100gbit/s DDoS (way too simplistic, but hopefully you get the idea) - way too small.
If you have the same traffic provisioned on average per customer and have 1 million customers, you can handle a DDoS 100x the size.
Only way to compete with this is to massively overprovision bandwidth per customer (which is expensive, as those customers won't pay more just for you to have more redundancy because you are smaller).
In a way (like many things in infrastructure) CDNs are natural monopolies. The bigger you get -> the more bandwidth and PoP you can have -> more attractive to more customers (this repeats over and over).
It was probably very astute of Cloudflare to realise that offering such a generous free plan was a key step in this.
There is no network protocol per se, but there is commercial solutions like fortinet that can block countries iirc, but to note that it's only ip range based so it's not worth a lot
edit: yes, you can you bgp to blockhole subnet traffic - the standard doesn't play well if you want blackhole unrelated subnets from upstream network
Blocking individual IP addresses? Sure, but consider that before your service detects enough anomalous traffic from one particular IP and is able to send the request to block upstream, your service will already be down from the aggregate traffic. Even a "slow" ddos with <10 packets per second from one source is enough to saturate your 10Gbps link if the attacker has a million machines to originate traffic from.
They have all the data on what CPE a user has, can send a letter and email with a deadline, and cut them off after it expires and the router has not been updated/is still exposed to the wide internet.
(Turns out some raspi reseller shipped a product with empty uname/password)
While a cute story, how do you scale that? And what about all the users that would be incapable of troubleshooting it, like if their laptop, roku, or smart lightbulb were compromised? They just lose internet?
And what about a botnet that doesn’t saturate your connection, how does your ISP even know? They get full access to your traffic for heuristics? What if it’s just one curl request per N seconds?
Not many good answers available if any.
Uh, yes. Exactly and plainly that. We also go and suspend people's driver licenses or at the very least seriously fine them if they misbehave on the road, including driving around with unsafe cars.
Access to the Internet should be a privilege, not a right. Maybe the resulting anger from widespread crackdowns would be enough of a push for legislators to demand better security from device vendors.
> And what about a botnet that doesn’t saturate your connection, how does your ISP even know?
In ye olde days providers had (to have to) abuse@ mailboxes. Credible evidence of malicious behavior reported to these did lead to customers getting told to clean up shop or else.
And even if the attack comes from your country, it is better to block part of the customers and figure out what to do next rather than have your site down.
Identifying and dynamically blocking the 500k offending IPs would certainly be possible technically -- 500k /32s is not a hard filtering problem -- but I seriously question the operational ability of internet providers to perform such granular blocking in real-time against dynamic targets.
I also have concerns that automated blocking protocols would be widely abused by bad actors who are able to engineer their way into the network at a carrier level (i.e. certain governments).
No, in my concept the host can only manage the traffic targeted at it and not at other hosts.
Is this really true? What device in the network are you loading that filter into? Is it even capable of handling the packet throughput of that many clients while also handling such a large block list?
But this would require a service like DNSBL / RBL which email providers use. Mutually trusting big players would exchange lists of IPs currently involved in DDoS attacks, and block them way downstream in their networks, a few hops from the originating machines. They could even notify the affected customers.
But this would require a lot of work to build, and a serious amount of care to operate correctly and efficiently. ISPs don't seem to have a monetary incentive to do that.
In a CDN, customers consume bandwidth; they do not contribute it. If Cloudflare adds 1 million free customers, they do not magically acquire 1 million extra pipes to the internet backbone. They acquire 1 million new liabilities that require more infrastructure investment.
All you are doing is echoing their pitch book. Of course they want to skim their share of the pie.
However most customers are rarely at their peak, this gives you tremendous spare capacity to use to eat DDoS attacks, assuming that the attacks are uncorrelated. This gives you huge amounts of capacity that's frequently doing nothing. Cloudflare advertise this spare capacity as "DDoS protection."
I suppose in theory it might be possible to massively optimise utilisation of your links, but that would be at the cost of DDoS protection and might not improve your margin very meaningfully, especially is customers care a lot about being online.
OP is saying it's cheaper overall for a 10 million customer company to add infrastructure for 1 million more than it is for a 10,000 customer company to add infrastructure for 1000 more people.
If you're looking at this as a "share of the pie", it's probably not going to make sense. The industry is not zero sum.
They contribute money which buys infrastructure.
> If Cloudflare adds 1 million free customers,
Is the free tier really customers? Regardless most of them are small that it doesn't cost cloudflare much anyways. The infrastructure is already there anyways. Its worth it to them for the good will it generates which leads to future paying customers. It probably also gives them visibility into what is good vs bad traffic.
1 million small sites could very well cost less to cloudflare than 1 big site.
The same reason I use cloud compute -- elastic infrastructure because I can't afford the peaks -- is the same reason large service providers "work".
It's funny how we always focus on Cloudflare, but all cloud providers have this same concentration downside. I think it's because Cloudflare loves to talk out of both sides of their mouth.
That's AI bots, BTW. Bots like Playwright or Crawl4AI, which provide a useful service to individuals using agentic AI. Cloudflare is hostile to these types of users, even though they likely cost websites nothing to support well.
The "scale saves money" argument commits a critical error: it counts only the benefits of concentration while externally distributing the costs.
Yes, economies of scale exist. But Cloudflare's scale creates catastrophic systemic risk that individual companies using cloud compute never would. An estimated $5-15 billion was lost for every hour of the outage according to Tom's Guide. That cost didn't disappear. It was transferred to millions of websites, businesses, and users who had zero choice in the matter.
Again, corporations shitting on free users. It's a bad habit and a dark pattern.
Even worse, were you hoping to call an Uber this morning for your $5K vacation? Good luck.
This is worse than pure economic inefficiency. Cloudflare operates as an authorized man-in-the-middle to 20% of the internet, decrypting and inspecting traffic flows. When their systems fail, not due to attacks, but to internal bugs in their monetization systems, they don't just lose uptime.
They create a security vulnerability where encrypted connections briefly lose their encryption guarantee. They've done this before (Cloudbleed), and they'll do it again. Stop pretending to have rational arguments with irrational future outcomes.
The deeper problem: compute, storage, and networking are cheap. The "we need Cloudflare's scale for DDoS protection" argument is a circular justification for the very concentration that makes DDoS attractive in the first place. In a fragmented internet with 10 CDNs, a successful DDoS on one affects 10% of users. In a Cloudflare-dependent internet, a DDoS, or a bug, affects 50%, if Cloudflare is unable to mitigate (or DDoSs themselves).
Cloudflare has inserted themselves as an unremovable chokepoint. Their business model depends on staying that chokepoint. Their argument for why they must stay a chokepoint is self-reinforcing. And every outage proves the model is rotten.
i don't even think they are evil because of the concentration of power, that's just a problematic issue. the evil part is they convince themselves they aren't the bad guys. that they are saving us from ourselves. that the things they do are net positives, or even absolute positives. like the whole "let's defend the internet from AI crawlers" position they appointed themselves sheriff on, that i think you're referencing. it's an extremely dangerous position we've allowed them to occupy.
> they monetize it
yes, and they can't do this without the scale.
> scale saves money
any company, uber for example, can design their infra to not rely on a sole provider. but why? their customers aren't going to leave in droves when a pretty reliable provider has the occasional hiccup. so it's not worth the cost, so why shouldn't they externalize it? uber isn't in business to make the internet a better place. so yes, scale does save money. you're arguing something at a higher principle than how architectural decisions are made.
i'm not defending economy of scale as a necessary evil. i'm just backing up that it's how cloudflare is built, and that it is in fact useful to customers.
Not every company can be an expert at everything.
But perhaps many of us could buy a different CDN than the major players if we want to reduce the likelihood of mass outages like this though.
Well that and the fact that when 99% goes through a central party, then that central party will be very interesting for authoritarian governments to apply sweeping censorship rules to.
Eventually?
It is not as bad as Cloudflare or AWS because certificates will not expire the instant there is an outage, but considers that:
- It serves about 2/3 of all websites
- TLS is becoming more and more critical over time. If certificates fail, the web may as well be down
- Certificate lifetimes are becoming shorter and shorter, now 90 days, but Let's Encrypt is now considering 6 days, with 47 days being planned as a minimum
- An outage is one thing, but should a compromise happen, that would be even more catastrophic
Let's Encrypt is a good guy now, but remember that Google used to be a good guy in the 2000s too!
Let’s Encrypt is great at making the existing system less painful, and there are a few alternatives like ZeroSSL, but all of this automation is basically a pile of workarounds on top of a fundamentally inappropriate design.
But DNSSEC was hard according to some, and now we are running a massive SPOF in terms of TLS certificates.
I'm also concerned about LE being a single point of failure for the internet! I really wish there were other free and open CAs out there. Our goal is to encrypt the web, not to perpetuate ourselves.
That said, I'm not sure the line of reasoning here really holds up? There's a big difference between this three-hour outage and the multi-day outage that would be necessary to prevent certificate renewal, even with 6-day certs. And there's an even bigger difference between this sort of network disruption and the kind of compromise that would be necessary to take LE out permanently.
So while yes, I share your fear about the internet-wide impact of total Let's Encrypt collapse, I don't think that these situations are particularly analogous.
It's genuinely insane that many companies are designing a great amount of fallbacks... on the software level but almost none is thought on the hardware/infrastructure level, common-sense dictate that you should never host everything on a single provider.
I was wrong, and ever since I've dealt with a targeted attack (which was evolving as I added more CF firewall rules). At this point it's taken care of, but only because I have most things completely blocked at the CF firewall layer.
(also congrats on 1 million subscribers but I know you must be tired of listening it but have a nice day jeff! Your videos are awesome!)
The idea that if companies like my former employer would stop doing DRM their audience would embrace it is rife idealism. But based on bitter experience so enough people will do bad things just for the lulz that you need to cover your ass.
My home lab will never have an open port, I'll always put things behind a CDN or zero trust system, even then...
FWIW, it's worthwhile just for educational reasons to look at abuseipdb.com quite revealing.
That being said, streaming content security is more than just DRM and DRM is more than just copy protection. There's a whole suite of tools inside DRM systems to manage content access at different levels and rulesets that can be applied for different situations. It's still fundamentally controlling an encrypted bitstream however. But I've implemented a great deal more than just DRM in order to build a better content security platform. Transit level controls, advanced token schemes, visible/invisible watermarking, threat/intrusion detection and abuse detection, there's quite a bit that can be implemented.
I was at Softlayer before I was at AWS and what catalyzed the move was the time I needed to add another hard drive to a system and somehow they screwed it up. I couldn't put a trouble ticket it to get it fixed because my database record in their trouble ticket system was corrupted. The next day I moved my stuff to AWS and the day after that they had a top sales guy talk to me to try to get me to stay but it was too late.
Then, with regular VPSs I also had systems down for 1-2 days. Just last week the company that hosts NextCloud for us was down the whole weekend (from Friday evening) and we couldn’t get their attention until Monday.
So far these huge outages that last 2-5 hours are still lower impact for me, and require me to take less action.
It depends on how you calculate your cost. If you only include the physical infrastructure having a dedicated server is cheaper. But by having some dedicated server you loose a lot of flexibility. Needs more resources? Just scale up your ec2, and with a dedicated server there is a lot more work involved.
Do you want a 'production-ready' database? With AWS you can just click a few buttons and have a RDS ready to use. To roll out your own PG installation you need someone with a lot of knowledge(how to configure replication? backups? updates? ...).
So if you include salaries in the calculation the result changes a lot. And even if you already have some experts in your payroll by putting them to work in deploying a PG instance you won't be able to use them to build other things that may generate more value to you business than the premium you pay to AWS.
"I added a load-balancer to improve system reliability" (happy)
"Load balancer crashed" (smiling-through-the-pain)
Technically, multi-node cluster with failover (or full on active-active) will have far higher uptime than just a single node.
Practically, to get the multi-node cluster (for any non trivial workload) to work right, reliably, fail-over in every case etc. is far more work, far more code (that can have more bugs), and even if you do everything right and test what you can, unexpected stuff can still kill it. Like recently we had uncorrectable memory error which just happened to hit the ceph daemon just right that one of the OSDs misbehaved and bogged down entire cluster...
Of course, if a big incident happens for a big CDN, there might not be enough latent capacity in the other CDNs to take all the traffic. CDNs are a cutthroat business, with small margins, so there usually isn’t a TON of unused capacity laying around.
So while us in tech might like a "snow day", there are millions of small businesses and people trying to go about their day to day lives who get cut off because of someone else's fuck-ups when this happens.
Even if you were making a million a minute, typically, it still didn't cost you a thing, nor have you lost anything.
You're not making as much, sure, but neither a cost, nor a loss.
Don't ask me how I know.
Maybe Tuesdays tend to be a big day for me, and instead of "down for a day", it's "lose almost a quarter of my income for the month".
Cloudflare is pretty pervasive, there are all kinds of people and businesses, in all kinds of situations, impacted by this.
But there are systems that depend on Cloudflare, directly or not, and when they go down it can have a serious impact on somebody's livelihood.
As as software engineer, I get it. as a CTO, I spent this morning triaging with my devops ai(actual Indian) to find some workaround (we found one) while our CEO was doing damage control with customers (non technical field) who were angry that we were down and they were losing business by the minute.
sometimes I miss not having a direct stake in the success of the business.
there was a point (maybe still) where not having a netflix subscription was seen as 'strange'.
if that's the case in your social circles -- and these kind of social things bother you -- you're not going to cancel the subscription due to bad service until it becomes a socially accepted norm.
If it can tell us that the host is up, surely it can just bypass itself to route traffic.
Congratulations, you've successfully completed Management Training 101.
It sucks there's not more competition in this space but CloudFlare isn't widely used for no reason.
AWS also solves real problems people have. Maintaining infrastructure is expensive as is hardware service and maintenance. Redundancy is even harder and more expensive. You can run a fairly inexpensive and performant system on AWS for years for the cost of a single co-located server.
It's like github vs whatever else you can do with git that is truly decentralized. The centralization has such massive benefits that I'm very happy to pay the price of "when it's down I can't work".
Very worrying indeed.
It's very frustrating of course, and it's the nature of the beast.
Not sure I follow, I didn't say it wasn't worrying or an issue. Just the reasons for it getting to this point are valid.
These decision are made individually not centrally. There is no process in place (and most likely there will never be) that will be able to control and dictate if people decide one way of doing things is the best way to do it. Even assuming they understand everything or know of the pitfalls.
Even if you can control individually what you do for the site you operate (or are involved in) you won't have any control on parts of your site (or business) that you rely on where others use AWS or Cloudflare.
AWS - someone touches DynamoDB and it kills the DNS.
Cloudflare - someone touches functionality completely unrelated to DNS hosting and proxying and, naturally, it kills the DNS.
There is this critical infrastructure that just becomes one small part of a wider product offering, worked on by many hands, and this critical infrastructure gets taken down by what is essentially a side-effect.
It's a strong argument to move to providers that just do one thing and do it well.
It has been dead to me since the SSL cache vulnerability thing and the arrogance with which senior people expected others to solve their problems.
But consider how many people still do stupid things like use the default CDN offered by some third party library, or use google fonts directly; people are lazy and don't care.
But I dare say the folks at these organisations take these matters incredibly seriously and the centralisation problem is largely one of risk efficiency.
I think there is no excuse, however, to not have multi region on state, and pilot light architectures just in case.
A lot (and I mean a lot) of people in IT like centralization specifically because it’s hard to blame people for doing something that everyone else is doing.
I'd be horrified. That's not the internet or computing industries I grew up with, or started working in.
But as long as the SPY keeps hitting > 10% returns each year, everyone's happy.
Perhaps the most graceful death of a tech company is that sentiment? Before some perception shift?
As always, in the name of "security". When are we going to learn that anything done, either by the government or by a corporation, in the name of security is always bad for the average person?
People not being ready for cloudflare/[insert hyperscaler] to be possibly down is the only fault.
And it’s hard to protect against DDoS without something like Cloudflare.
Look at the posts here.
Even the meager HN “hug of death” will take things down
Companies can always do as they please and people will rationalize anything.
Dialogue about mitigations/solutions? Alternative services? High availability strategies?
Nah! It's free to complain.
Me personally, I'd say those companies do a phenomenal job by being a de facto backbone of the modern web. Also Cloudflare, in particular, gives me a lot of things for free.
The target these days is the user.
The make-believe worm.
Your power is provided by a power utility company. They usually serve an entire state, if not more than one (there are smaller ones too). That's "centralization" in that it's one company, and if they "go down", so do a lot of businesses. But actually it's not "centralized", in that 1) there are actually many different companies across the country/world, and 2) each company "decentralizes" most of its infrastructure to prevent massive outages.
And yes, power utilities have outages. But usually they are limited in scope and short-lived. They're so limited that most people don't notice when they happen, unless it's a giant weather system. Then if it's a (rare) large enough impact, people will say "we need to reform the power grid!". But later when they've calmed down, they realize that would be difficult to do without making things worse, and this event isn't common.
Large internet service providers like AWS, Cloudflare, etc, are basically internet utilities. Yes they are large, like power utilities. Yes they have outages, like power utilities. But the fact that a lot of the country uses them, isn't any worse than a lot of the country using a particular power company. And unlike the power companies, we're not really that dependent on internet service providers. You can't really change your power company; you can change an internet service provider.
Power didn't used to be as reliable as it is. Everything we have is incredibly new and modern. And as time has passed, we have learned how to deal with failures. Safety and reliability has increased throughout critical industries as we have learned to adapt to failures. But that doesn't mean there won't be failures, or that we can avoid them all.
We also have the freedom to architect our technology to work around outages. All the outages you have heard about recently could be worked around, if the people who built on them had tried:
- CDN goes down? Most people don't absolutely need a CDN. Point your DNS at your origins until the CDN comes back. (And obviously, your DNS provider shouldn't be the same as your CDN...)
- The control plane goes down on dynamic cloud APIs? Enable a "limp mode" that persists existing infrastructure to serve your core needs. You should be able to service most (if not all) of your business needs without constantly calling a control plane.
- An AZ or region goes down? Use your disaster recovery plan: deploy infrastructure-as-code into another region or AZ. Destroy it when the az/region comes back.
...and all of that just to avoid a few hours of downtime per year? It's likely cheaper to just take the downtime. But that doesn't stop people from piling on when things go wrong, questioning whether the existence of a utility is a good idea.
Are people really this confused?